A live editorial feed of product news, API changes, availability checks, pricing signals, comparison refreshes and new Chinese AI guides.
Freshness answer
Product updates · Release
Z.ai released GLM-5.3 for all GLM Coding Plan users, using the same base model as GLM-5.2 with gains driven by scaled post-training. The vendor reports a 50% improvement on its private Z.ai Code Bench, stronger long-horizon coding and major cybersecurity gains. GLM-5.3 requires thinking to remain enabled with low, high or max reasoning effort. The general API is still marked coming soon, and weights are planned two weeks after launch following safety evaluation and hardening.
Product updates · Open source
DeepSeek released DeepSeek Harness as an MIT-licensed, local-first coding agent and agent development/runtime environment. Built on Cordis, it implements models, tools, skills, sessions, sandboxes, storage, loops, scheduling and UI as configurable plugins; records append-only trajectories for inspection and replay; and ships Standard, Code, Minimal and Creator modes. The project remains in developer preview and explicitly warns of compatibility-breaking changes.
Availability check
Globally available · Full English UI · 5/5 sources verified, recently checked
Availability check
Partially available · Partial English UI · 8/8 sources verified, recently checked
Availability check
Globally available · Full English UI · 4/4 sources verified, recently checked
Availability check
Globally available · Full English UI · 4/4 sources verified, recently checked
Product updates · Release
AgentScope AI completed the public rename from HiClaw to AgentTeams and released v1.2.0 stable. The current repository standardizes Team and Worker contracts across container images, Kubernetes APIs, Helm, Matrix, storage and runtimes, while adding an optional dashboard and retaining OpenClaw, QwenPaw and Hermes collaboration.
Product updates · Open source
The official openPangu Hugging Face organization now hosts openPangu-2.0-Flash and Flash-Int8, published June 30, plus openPangu-2.0-Pro and Pro-Int8, published July 30. The repositories expose downloadable safetensors and custom model code under a custom model license, resolving the original announcement-stage availability gap while leaving hardware portability and independent benchmark reproduction open.
Product updates · Open source
Moonshot AI published Kimi K3's full weights, model code and custom model license on Hugging Face on July 27. The public repository contains an MXFP4-compressed checkpoint split across 96 weight shards, closing the launch announcement's open-weight commitment while leaving substantial infrastructure and license-review requirements for self-hosting.
Product updates · Pricing
MiniMax's live API pricing lists permanently discounted M3 rates for context up to 512K at $0.30/M input, $1.20/M output and $0.06/M cache-read tokens; 512K-1M costs $0.60, $2.40 and $0.12, with access above 512K currently limited. M2.7 costs $0.30/M multimodal input and $1.20/M output, while Highspeed costs $0.60 and $2.40. Hailuo 2.3 starts at $0.28 per 768P six-second video, Speech 2.8 Turbo at $60/M characters, rapid voice cloning at $1.50 per voice, voice design at $3 per voice, and Music 2.6 at $0.15 per song with a free-trial label.
Availability check
Globally available · Full English UI · 4/4 sources verified, recently checked
Availability check
Partially available · Full English UI · 4/4 sources verified, recently checked
Availability check
Partially available · Full English UI · 3/3 sources verified, recently checked
Availability check
Globally available · Full English UI · 4/4 sources verified, recently checked
Availability check
Globally available · Full English UI · 3/3 sources verified, recently checked
Availability check
Globally available · Full English UI · 2/2 sources verified, recently checked
Availability check
Partially available · Partial English UI · 3/3 sources verified, recently checked
Availability check
Partially available · Partial English UI · 3/3 sources verified, recently checked
Availability check
Globally available · Full English UI · 3/3 sources verified, recently checked
Availability check
Globally available · Full English UI · 8/8 sources verified, recently checked
Availability check
Partially available · Full English UI · 5/5 sources verified, recently checked
Availability check
Globally available · Full English UI · 6/6 sources verified, recently checked
Availability check
Globally available · Full English UI · 6/6 sources verified, recently checked
Availability check
Partially available · Full English UI · 5/5 sources verified, recently checked
Availability check
Globally available · Full English UI · 4/4 sources verified, recently checked
Availability check
Globally available · Full English UI · 5/5 sources verified, recently checked
Availability check
Globally available · Partial English UI · 10/10 sources verified, recently checked
Availability check
Globally available · Full English UI · 2/2 sources verified, recently checked
Availability check
Globally available · Full English UI · 5/5 sources verified, recently checked
Availability check
Globally available · Full English UI · 2/2 sources verified, recently checked
Availability check
Globally available · Full English UI · 1/1 sources verified
Availability check
Globally available · Full English UI · 1/1 sources verified
Availability check
Globally available · Full English UI · 2/2 sources verified, recently checked
Availability check
Globally available · Full English UI · 6/6 sources verified, recently checked
Availability check
Globally available · Full English UI · 3/3 sources verified, recently checked
Availability check
Globally available · Full English UI · 3/3 sources verified, recently checked
Availability check
Globally available · Full English UI · 2/2 sources verified, recently checked
Availability check
Globally available · Full English UI · 3/3 sources verified, recently checked
Availability check
Globally available · Full English UI · 3/3 sources verified, recently checked
Availability check
Globally available · Full English UI · 3/3 sources verified, recently checked
Availability check
Globally available · Full English UI · 5/5 sources verified, recently checked
Availability check
Globally available · Full English UI · 5/5 sources verified, recently checked
Availability check
Globally available · Full English UI · 6/6 sources verified, recently checked
Availability check
Globally available · Full English UI · 5/5 sources verified, recently checked
Availability check
Globally available · Full English UI · 6/6 sources verified, recently checked
Availability check
Globally available · Full English UI · 4/4 sources verified, recently checked
Availability check
Globally available · Full English UI · 4/4 sources verified, recently checked
Availability check
Globally available · Full English UI · 7/7 sources verified, recently checked
Availability check
Partially available · Full English UI · 3/3 sources verified, recently checked
Availability check
Partially available · Full English UI · 5/5 sources verified, recently checked
Availability check
Globally available · Full English UI · 1/1 sources verified
Availability check
Partially available · Full English UI · 1/1 sources verified
Availability check
Partially available · Partial English UI · 2/2 sources verified, recently checked
Availability check
Partially available · Partial English UI · 1/1 sources verified
Availability check
Partially available · Full English UI · 1/1 sources verified
Availability check
Partially available · Full English UI · 1/1 sources verified
Availability check
Globally available · Partial English UI · 1/1 sources verified
Availability check
Unknown · Partial English UI · 1/1 sources verified
Availability check
Unknown · Partial English UI · 1/1 sources verified
Availability check
Partially available · Partial English UI · 2/2 sources verified, recently checked
Availability check
Partially available · Partial English UI · 2/2 sources verified, recently checked
Availability check
Partially available · Partial English UI · 2/2 sources verified, recently checked
Availability check
Partially available · Full English UI · 3/3 sources verified, recently checked
Availability check
Partially available · Full English UI · 2/2 sources verified, recently checked
Availability check
Partially available · Full English UI · 2/2 sources verified, recently checked
Availability check
Globally available · Full English UI · 2/2 sources verified, recently checked
Availability check
Globally available · Full English UI · 7/7 sources verified, recently checked
Availability check
Globally available · Full English UI · 3/3 sources verified, recently checked
Availability check
Globally available · Full English UI · 6/6 sources verified, recently checked
Availability check
Globally available · Full English UI · 2/2 sources verified, recently checked
Availability check
Globally available · Full English UI · 5/5 sources verified, recently checked
Availability check
Globally available · Full English UI · 5/5 sources verified, recently checked
Availability check
Globally available · Full English UI · 1/1 sources verified
Availability check
Partially available · Full English UI · 2/2 sources verified, recently checked
Availability check
Partially available · Partial English UI · 4/4 sources verified, recently checked
Availability check
Partially available · Partial English UI · 2/2 sources verified, recently checked
Availability check
Partially available · Full English UI · 5/5 sources verified, recently checked
Availability check
Globally available · Full English UI · 2/2 sources verified, recently checked
Availability check
Globally available · Full English UI · 4/4 sources verified, recently checked
Availability check
Globally available · Partial English UI · 4/4 sources verified, recently checked
Availability check
Partially available · Full English UI · 1/1 sources verified
Availability check
Partially available · Full English UI · 2/2 sources verified, recently checked
Availability check
Partially available · Full English UI · 4/4 sources verified, recently checked
Availability check
Partially available · Full English UI · 3/3 sources verified, recently checked
Availability check
Partially available · Full English UI · 1/1 sources verified
Availability check
Partially available · Full English UI · 1/1 sources verified
Availability check
Partially available · Full English UI · 3/3 sources verified, recently checked
Availability check
Partially available · Full English UI · 2/2 sources verified, recently checked
Availability check
Globally available · Full English UI · 2/2 sources verified, recently checked
Availability check
Globally available · Full English UI · 1/1 sources verified
Availability check
Globally available · Full English UI · 4/4 sources verified, recently checked
Availability check
Globally available · Full English UI · 3/3 sources verified, recently checked
Availability check
Globally available · Full English UI · 3/3 sources verified, recently checked
Availability check
Globally available · Partial English UI · 2/2 sources verified, recently checked
Availability check
Globally available · Full English UI · 3/3 sources verified, recently checked
Availability check
Globally available · Full English UI · 2/2 sources verified, recently checked
Availability check
Partially available · Full English UI · 4/4 sources verified, recently checked
Availability check
Partially available · Partial English UI · 1/1 sources verified
Availability check
Partially available · Full English UI · 2/2 sources verified, recently checked
Availability check
Partially available · Full English UI · 5/5 sources verified, recently checked
Availability check
Partially available · Full English UI · 4/4 sources verified, recently checked
Availability check
Partially available · Full English UI · 4/4 sources verified, recently checked
Availability check
Partially available · Full English UI · 3/3 sources verified, recently checked
Availability check
Partially available · Full English UI · 2/2 sources verified, recently checked
Availability check
Partially available · Full English UI · 1/1 sources verified
Availability check
Globally available · Full English UI · 5/5 sources verified, recently checked
Availability check
Globally available · Full English UI · 4/4 sources verified, recently checked
Availability check
Partially available · Full English UI · 2/2 sources verified, recently checked
Availability check
Partially available · Full English UI · 2/2 sources verified, recently checked
Availability check
Globally available · Full English UI · 3/3 sources verified, recently checked
Availability check
Globally available · Full English UI · 4/4 sources verified, recently checked
Availability check
Globally available · Full English UI · 2/2 sources verified, recently checked
Availability check
Partially available · Full English UI · 4/4 sources verified, recently checked
Availability check
Globally available · Full English UI · 1/1 sources verified
Availability check
China only · No English UI · 1/1 sources verified
Comparison update
Choose Kling when you specifically want to evaluate Chinese video models and international access is confirmed. Choose Runway when enterprise workflow maturity is the priority.
Comparison update
Use Google Gemini when you need a broad multimodal assistant for mixed reasoning and productivity workflows. Use Seedance 2.0 when the deliverable is generated video and direct motion-oriented creative control matters.
Product updates · Release
Moonshot AI launched Kimi K3 as a hosted 2.8T-parameter, native-vision MoE model with a 1M-token context window. It became available in Kimi, Kimi Work, Kimi Code and the Kimi API as kimi-k3 at $0.30/M cache-hit input tokens, $3/M cache-miss input tokens and $15/M output tokens. At launch, Moonshot scheduled the full-weight release for July 27; that release is now tracked separately. Published limitations include sensitivity to dropped thinking history and possible over-proactiveness on ambiguous tasks.
Product updates · Availability
Huawei Cloud's international Pangu page now anchors the catalog profile with an English product, sign-up, console, documentation and contact-sales path. It defines Pangu's three-layer ToB architecture across foundation, industry-tailored and scenario-specific models; lists NLP, CV, scientific-computing, multimodal and prediction foundation models; and positions ModelArts Studio as the access layer for inference, data engineering, model development, industry applications and agents. The profile also records the AP-Southeast-1 console route, enterprise pricing uncertainty and examples spanning digital employees, certificate review, smart cities, coal blending and weather forecasting.
Product updates · Release
OpenRouter's Owl Alpha is now tracked as a free text model for agentic coding, tool use, automated workflows and complex instruction execution. The live model API lists a 1,048,756-token context window, up to 262,144 output tokens, native tools and structured outputs, zero input and output token prices, and one Stealth INT8 provider route. OpenRouter says it works with Claude Code and OpenClaw, but also warns that the provider may log prompts and completions and use them for model improvement. The underlying model identity and license remain undisclosed, so the profile treats it as a non-sensitive Alpha evaluation route rather than a production-safe default.
Product updates · Open source
DeepReinforce's Hugging Face organization now exposes the Ornith 1.0 family of MIT-licensed agentic-coding models. The collection includes public 9B, 35B and 397B model cards plus GGUF and FP8 variants. The cards position Ornith as a self-improving model family post-trained from Gemma 4 and Qwen 3.5, using reinforcement learning to optimize both solution rollouts and their scaffolds, with reported results on Terminal-Bench 2.1, SWE-Bench, NL2Repo, SWE Atlas and ClawEval. The profile records vLLM, SGLang, Transformers and OpenAI-compatible local serving paths.
Product updates · Open source
Qwen has released Qwen-AgentWorld-35B-A3B open weights and the AgentWorldBench evaluation benchmark. The project positions Qwen-AgentWorld as a native language world model for simulating agentic environments across MCP, Search, Terminal, SWE, Android, Web and OS domains. The README says the 35B MoE model has 3B active parameters and a 256K context, can be served through SGLang or vLLM with an OpenAI-compatible API, and that AgentWorldBench scores predicted environment observations on format, factuality, consistency, realism and quality. The repository also reports 397B research results, but only the 35B-A3B weights are listed as the open-source release.
Product updates · API
BytePlus now exposes OmniHuman 1.5 as a dedicated image-driven avatar video model page. The page says it creates videos from a single image and multimodal prompts covering audio, image and text, supports native 1080p output, rhythmic and emotional performances, multi-person scenes, gesture control and camera movement. Public links include Access API, API docs, a prompt guide, a tech report, product brochure and Discord. The visible page lists pricing at $0.12 per second, while page metadata still says $0.16 per second, so checkout pricing should be rechecked before purchase.
Product updates · Pricing
BytePlus now exposes ArkClaw as a dedicated AI-agent product page. It is positioned as a 24/7 AI partner with plug-and-play setup, multi-agent collaboration, private secure environments, Lark-native collaboration, 50+ built-in Skills and at least 40 GB of dedicated persistent storage. The public page lists a Dedicated ArkClaw Assistant from $10/month for the first month and $22/month from the second month, plus an ArkClaw Combined Package from $20/month for the first month and $32/month from the second month that bundles ModelArk Coding Plan LLM quota. Checkout options, quota details and regional availability still need account verification.
Product updates · Pricing
BytePlus now exposes a dedicated Seedream 5.0 Lite product page with ModelArk vision experience links, API documentation, free quota and pricing. The page positions Seedream 5.0 Lite as a next-generation image model with stronger reasoning and comprehension, better global knowledge, reference consistency, professional scene generation, character consistency and feature preservation. It lists $0.035 per image, 50 free image generations, modelId seedream-5-0-260128, API docs, a prompt brochure and first-purchase packs for 400, 1028 and 2000 images valid for 30 days.
Product updates · Release
ByteDance Seed has released Seed2.1 with Pro and Turbo variants for real-world productivity. The official page emphasizes stronger general-agent execution and code engineering, covering project planning, file processing, tool use, requirement understanding, coding, debugging and validation. It also reports gains in knowledge, reasoning, visual understanding, spatial reasoning, long-context processing and video understanding, including hour-long video processing and SOTA results across several video benchmarks. Seed2.1 links to Try it and API entry points, but production model availability and quotas still need console verification.
Product updates · Market
Microsoft Partner Center says Copilot Cowork became generally available on June 16, 2026 as an agentic system for complex, long-running, multi-tool work inside Microsoft 365. Cowork requires a Microsoft 365 Copilot license and usage-based billing with Copilot Credits, with admin controls for access, budgets and usage. Microsoft lists multiple models, security and compliance capabilities, partner and Dynamics 365 plugins, and a June 30 billing setup deadline for Frontier customers. Axios separately reports Microsoft is considering a Microsoft-hosted DeepSeek option to lower Cowork model costs; this is recorded as a media-reported consideration, not an official model launch.
Product updates · Pricing
BytePlus now lists Dreamina Seedance 2.0 resource plans for international users: Light at $30.10 for 7M tokens, Production at $43.00 for 10M tokens and Premium at $55.90 for 13M tokens, each valid for three months. The plans advertise about 28, 40 and 52 480P videos respectively, support 480P, 720P and 1080P output, 4-15 second duration and multimodal video generation. Purchased packages cannot be canceled or refunded, cannot be offset by the AI savings plan and fall back to pay-as-you-go billing after exhaustion.
Product updates · API
Qwen Cloud now lists HappyHorse-1.1-T2V as the current text-to-video model, with improved semantic understanding, cinematic shot control, dynamic motion rendering, richer details and stronger visual consistency. The model uses the asynchronous DashScope video-synthesis endpoint, supports 720P and 1080P output, advertises 40%-off rates of $0.084/second and $0.108/second, and lists 300 RPM, five concurrent requests, a 500-task async queue and ten free seconds. The public DingTalk document title echoes the release theme, but its body text was not publicly extractable.
Product updates · Release
MiniMax Speech 2.8 adds native sound tags for breaths, pauses, hesitations and expressive actions, plus higher-fidelity voice cloning from a 10-second sample and cleaner output with reduced background noise and digital artifacts. MiniMax also highlights improved cross-lingual pronunciation and reduced accent bleed for the Mandarin-Japanese pair, with more languages planned. The model is live through MiniMax Audio and the Open Platform API; cloning quality and human-indistinguishability claims remain vendor-reported and require consent-aware testing.
Product updates · Release
MiniMax Music 2.6 adds stronger rhythmic tension and pauses, reworked low-end acoustics, lo-fi and indie vocal character, prompt control over BPM, key, song structure and emotional arc, and a new Cover workflow that preserves a source melody while changing genre, arrangement or lyrics. MiniMax reports first-packet latency below 20 seconds and released three Agent Skills: minimax-music-gen, minimax-music-playlist and buddy-sings. The announcement's 14-day beta benefits, including 500 consumer creations and 100 extra Token Plan API calls per day, were launch-period offers and are no longer presented as current entitlements.
Product updates · Release
MiniMax Hailuo 2.3 improves complex body motion, physics and command following, dynamic-camera rendering, stylization, facial micro-expressions and object-motion response over Hailuo 02. MiniMax says it retained Hailuo 02 pricing, while Hailuo 2.3 Fast can reduce batch-generation cost by up to 50%. The update is available through Hailuo AI web, mobile and the Open Platform API. Hailuo's former Video Agent also evolved into a globally launched Media Agent for one-click or step-by-step multimodal creation using text, images, video and audio.
Product updates · Pricing
Kimi API is running a K2.7 Code launch top-up rebate from June 11, 2026 at 09:00 PDT through July 2 at 08:59:59 PDT. A single top-up of $100-$299 receives a 20% voucher, $300-$999 receives 25%, and $1,000 or more receives 30%. Each Organization ID can participate once; the voucher is issued within one business day, capped at $4,000, valid for 90 days and usable across Kimi platform purchases. Top-up funds are non-refundable.
Product updates · Pricing
Kimi API's model-pricing overview confirms that Chat Completions bills both input and output tokens. Document content is also billed as input when extracted text is sent to a model, while file storage and file-content extraction interfaces are temporarily free. Exact usage can be checked through the Token Calculation API; model-specific rates remain on the detailed K2.7 Code, K2.6 and Moonshot V1 pricing pages.
Product updates · API
Kimi's K2.7 Code quickstart documents the kimi-k2.7-code and kimi-k2.7-code-highspeed model IDs, a 256K context window and a default 32K maximum output. HighSpeed targets about 180 tokens per second and up to 260 tokens per second for short contexts, though capacity is currently limited. Thinking cannot be disabled; temperature, top_p, n and penalty values are fixed, tool_choice only accepts auto or none, and multi-step tool calls must preserve reasoning_content. The OpenAI-compatible API also supports image and video content in multimodal tool results.
Product updates · Pricing
BytePlus ModelArk Coding Plan offers Lite and Pro subscriptions for supported coding tools, with monthly, quarterly and annual billing. It routes Dola-Seed, GLM, DeepSeek, Kimi and GPT-OSS models across Claude Code, Cursor, Cline, Kilo Code, Roo Code and OpenCode, with OpenClaw and Hermes Agent integrations. Lite is marketed as three times Claude Pro usage, while Pro provides five times Lite capacity and includes ArkClaw during the subscription period. Public numeric prices did not load in the signed-out page and should be verified at checkout.
Product updates · Pricing
MiniMax Token Plan lists Plus, Max and Ultra at $20, $50 and $120 per month, or $200, $500 and $1,200 per year with two months free. The tiers estimate 1.7B, 5.1B and 12.5B M3 tokens per month and 3-4, 4-5 and 6-7 concurrent agents. Quota is shared across supported modalities, governed by rolling five-hour and weekly windows, and does not carry over; MiniMax recommends pay-as-you-go billing for production workloads.
Product updates · Pricing
DeepSeek's current pricing lists V4-Flash at $0.0028/M cache-hit input, $0.14/M cache-miss input and $0.28/M output, with 2,500 concurrency. V4-Pro costs $0.003625, $0.435 and $0.87 respectively, with 500 concurrency. Both expose 1M context and 384K maximum output. The former deepseek-chat and deepseek-reasoner aliases have passed their July 24 deprecation date and are no longer listed on the current pricing page; peak/off-peak pricing is also planned for a date to be announced.
Product updates · Pricing
Qwen Cloud Token Plan now records Standard at $30/seat/month with 25,000 credits, Pro at $100 with 100,000 credits and Max at $200 with 250,000 credits. Seats can be bought in bulk and assigned to team members, and quota is shared across supported interactive tools. Team subscriptions are non-refundable; automated scripts and application backends are prohibited, and Token Plan Team keys cannot be mixed with Coding Plan or pay-as-you-go keys.
Product updates · Market
StepFun's official site now unifies its model and product entry points with an English view, Step 3.7 Flash as the headline fully open native-vision agent model, Step API, Step Plan and Studio links, plus Step AI clients for macOS 11.3+, Windows 10+, iOS, Android and the web.
Product updates · API
Qwen3.7-Max is now reflected as Qwen Cloud's largest Qwen3.7 model for text-only coding, productivity and long-horizon autonomous tasks. It exposes a 1M-token context, 65.53K maximum output, 600 RPM, 1M TPM, function calling, structured outputs, caching, batches and web search. The closed model page currently advertises 50%-off API rates of $1.25/M input and $3.75/M output tokens.
Product updates · API
Qwen Cloud's HappyHorse profile now covers text-to-video, image-to-video, reference-to-video and natural-language video editing. HappyHorse-1.0-T2V uses an asynchronous DashScope endpoint with 600 RPM, five concurrent requests and a 500-task queue; its current promotional rates are $0.112/second for 720P and $0.192/second for 1080P generation.
Product updates · API
Qwen3.7-Plus is now reflected in the Qwen Cloud profile with image, text and video input, text output, a 1M-token context, 65.53K maximum output, 15K RPM and 5M TPM limits. The closed model supports function calling, structured outputs, caching, batches, web search and visual-agent workflows; its model page shows promotional prices from $0.32-$0.96/M input and $1.28-$3.84/M output tokens depending on context length.
Product updates · Release
Qwen Code's profile now records the official installer, interactive terminal mode with @ file references, qwen -p headless automation, OpenAI/Anthropic/Gemini-compatible APIs, Qwen OAuth, Skills, SubAgents, VS Code/Zed/JetBrains integrations and the @qwen-code/sdk package for custom agent development.
Product updates · Market
GizmoChina reports that GLM-5.2 took the top position on the crowdsourced web-design benchmark Design Arena ahead of Claude Fable 5. VentureBeat separately cites a 1360 ELO score. This is recorded as a media-reported arena result, not an independently reproduced evaluation, and rankings may change as new votes and models are added.
Product updates · Market
VentureBeat reports GLM-5.2 API rates of $1.40 per million input tokens, $4.40 per million output tokens and $0.26 per million cached-input tokens. Its 'one-sixth the cost' headline compares the $5.80 input-plus-output total with GPT-5.5's listed $35 total; real savings depend on workload token mix, caching and provider terms. The article's benchmark comparisons largely reproduce Z.ai's vendor-reported results.
Product updates · Open source
Z.ai released GLM-5.2 with a stable 1M-token context, High and Max thinking-effort controls, MIT-licensed open weights and local support through Transformers, vLLM, SGLang, xLLM and KTransformers. It is available through Z.ai and GLM Coding Plan; Z.ai reports major gains over GLM-5.1 on long-horizon and coding benchmarks, though those benchmark claims remain vendor-reported.
Product updates · Pricing
Qoder's pricing now records Community free, Pro at $20/month with 2,000 credits, Pro+ at $60/month with 6,000 credits, Ultra at $200/month with 20,000 credits and Teams at $40/seat/month with 3,000 credits per seat. New users receive a one-time 14-day Pro trial with 300 credits; monthly credits expire, while paid users can buy extra Credit Packs.
Product updates · Open source
Qwen's official research index currently exposes 60 release, open-source and research entries across language models, coding, vision, image generation and editing, speech, omni-modal systems, embeddings, agents, safety and training methods. It is now used as a primary source for tracking the wider Qwen model family.
Product updates · Availability
Qoder Mobile is now tracked as Qoder's phone and mobile-web control surface for Desktop Quest, CLI and cloud tasks. It provides real-time status, push notifications, plan review, approval and follow-up controls while execution remains on the user's computer or in Qoder's cloud environment.
Product updates · Release
Qoder CLI is now tracked as Qoder's terminal coding agent with interactive and headless modes, tool allowlists, Git worktrees, AGENTS.md memory, GitHub Action workflows, managed cloud execution and TypeScript/Python Agent SDKs for application and CI integration.
Product updates · Release
Qoder's core profile now reflects the Desktop product's Editor and Quest workspaces, NEXT and Inline Chat, Agent and Experts delegation, 100,000-file indexing, Browser Agent, Knowledge Engine, review and PR workflows, current system support and public individual-plan pricing.
Product updates · API
Qoder Cloud Agents is now tracked as a managed Agent API with isolated container environments, persistent SSE sessions, tools and Skills, allow/ask/deny permission policies, encrypted credential Vaults, versioned Memory Stores and coordinator-led multi-agent execution.
Product updates · Release
QoderWake is now tracked as Qoder's locally managed, always-on digital-employee runtime. Its Wakers combine role templates, configurable memory and guardrails, scheduled, GitHub-event and API-triggered tasks, plus Leader-led employee groups for software engineering, product, data and content workflows.
Product updates · API
MiniMax M3's profile now includes the official model page's guaranteed 512K and maximum 1M API context, 83.5 BrowseComp score, long-horizon paper-reproduction and CUDA-optimization demonstrations, MiniMax-M3 API identifier and the remaining uncertainty around full weight publication.
Product updates · API
Kimi Code now documents how members can use its coding benefits in Claude Code through an Anthropic-compatible endpoint and in Roo Code through an OpenAI-compatible endpoint. The guide also names OpenCode, OpenClaw and Hermes as compatible tool or framework paths, and notes that K2.7 Code requires Thinking mode while non-thinking requests route to K2.6.
Product updates · Open source
JD's Joy Future Academy has released JoyAI-VL-Interaction as an open real-time video-language interaction system. The repository includes an 8B vision-first interaction model, model weights, training recipe, more than 4M time-aligned interaction samples, a deployable streaming stack, English and Chinese docs, Hugging Face model and dataset links, and day-0 vLLM-Omni deployment support. The model is designed to continuously watch webcam or livestream input and decide when to speak, stay silent or delegate to a background model/API/agent.
Product updates · Release
MiniMax Hub is now tracked as MiniMax's local-first multimodal creative command center, combining copy, image, video, audio, packaging, export, Skills, local asset management and multi-agent orchestration for short-form video, brand marketing and production workflows.
Product updates · Open source
NetEase Youdao's Confucius4-TTS is now tracked in AI Audio as an Apache-2.0 open-source multilingual and cross-lingual zero-shot TTS engine. The profile records its speech-encoder-plus-LLM architecture, 14 supported languages, unconstrained voice cloning without reference transcripts, cross-lingual voice transfer, emotion transfer, online demo, Hugging Face and ModelScope model paths, Python inference, fine-tuning workflow and benchmark sources.
Product updates · Open source
Kimi K2.7 Code is now tracked as Moonshot AI's open-weight coding-focused agentic model, with 1T total parameters, 32B activated parameters, 256K context, native multimodal input, forced thinking mode, Kimi Code default-model access and Kimi API pricing.
Product updates · Release
ZCode is now tracked as Z.ai's desktop ADE for long-horizon tasks. Version 3.0.0 switches to the in-house ZCode Agent, deepens GLM optimization and adds grouped multi-agent workspaces, Zread, Git branch graphs, usage monitoring and remote task continuation.
Product updates · Pricing
Z.ai's Lite, Pro and Max coding subscriptions are now tracked with standard monthly prices of $18, $72 and $160, annual discounts, estimated 80/400/1,600 prompts per five hours, weekly caps, GLM model multipliers, 20+ supported tools and MCP quotas.
Product updates · Open source
Kimi Code is now tracked from Moonshot AI's MIT-licensed repository and version 0.14.2, covering single-binary installers, video input, conversational MCP setup, trusted plugins, coder/explore/plan subagents, lifecycle hooks and ACP editor integration.
Product updates · Open source
HDC 2026 reports describe 505B/18B Pro and 92B/6B Flash models with 512K context and an Ascend-first design. Huawei plans to begin opening seven component groups on June 30; weights and final license terms are not yet public.
News
The public weights make openPangu 2.0 testable, but practical openness still depends on its custom model license, Ascend-first runtime requirements and independent reproducible evaluation.
Product updates · Open source
MiMoCode is now tracked as an MIT-licensed, OpenCode-based terminal agent with cross-session memory, context reconstruction, parallel subagents, goal judging, Compose workflows, voice input and custom model providers. Its separate use restrictions still apply.
News
The ranking is a strong signal of developer adoption and price-performance demand on OpenRouter, but it should not be interpreted as global market share across direct APIs, cloud platforms and consumer products.
Product updates · Pricing
OpenCode Go now has a refreshed profile focused on the official $10/month subscription, $5 first month offer, $12 / 5-hour, $30 / weekly and $60 / monthly usage limits, current model list and Go provider API setup.
Product updates · Release
MiniMax is now tracked with a dedicated M3 model profile, while the homepage currently highlights M3, Hailuo 2.3, Speech 2.8 and Music 2.6 alongside the API Platform, Token Plan and Agent surfaces.
Product updates · Market
Miora is now tracked as an agentic creative studio, with the public homepage currently positioning it around multi-agent visual, video, 3D and prototype workflows and linking out to CodeBuddy, WorkBuddy, Genie and Tencent Cloud.
Product updates · Market
WorkBuddy is now tracked as Tencent Cloud's international AI agent workspace, with the homepage currently exposing Hy3 preview, Kimi 2.6, MiniMax 2.7, GLM-5.1 and DeepSeek-V3.2 alongside public Free, Pro and Team pricing and office-app connectors.
Product updates · Release
Step 3.7 Flash is now tracked as StepFun's flagship multimodal reasoning model, with 196B / 11B sparse MoE architecture, 256K context, native image and video understanding, tool calling, reasoning-effort controls and public pricing.
Product updates · Release
MiniMax Agent's Mavis upgrade introduces parallel role-based Agent Teams for long-running work, background execution, verification, retries, recoverable task state, memory and Skills. MiniMax also merged Agent Plan and Token Plan so one subscription can share credits across Agent, CLI and API usage.
Product updates · Market
Skywork is now tracked as an AI workspace with public help and app surfaces for Images, Documents, Slides, Sheets, Websites, Videos, Tools, SkyClaw, SkyBot and AI Super Agents.
Product updates · Release
OpenCode Go is now tracked as a low-cost subscription for open coding models, with a $5 first month, $10/month after that, a beta model list that includes GLM, Kimi, MiMo, MiniMax, Qwen and DeepSeek, and provider docs for OpenCode Zen, API keys and usage limits.
Product updates · Release
SenseNova is now tracked as SenseTime's multimodal model platform, with the official models page highlighting SenseNova 6.7 Flash-Lite and SenseNova U1 Fast and the Token Plan page exposing free public-beta access, Hermes Agent and OpenClaw support.
Product updates · API
Alibaba Cloud Model Studio is now tracked as Alibaba Cloud's model platform for Qwen and third-party LLMs, with the homepage currently highlighting Qwen3.7-Max, Qwen3.6-Plus, HappyHorse 1.0 and Qwen-Plus alongside OpenAI-compatible APIs, API-key setup, regional deployment modes and migration paths.
News
Do not use the reported timing, quality uplift or unchanged-pricing claim in current buying guidance until ByteDance publishes an official model page.
Product updates · Open source
Qwen Code is now tracked in AI Coding as Qwen's terminal-first coding agent, covering Skills, SubAgents, Qwen OAuth, OpenAI/Anthropic/Gemini-compatible clients, headless mode and editor integrations for VS Code, Zed and JetBrains.
Product updates · Open source
QwenPaw is now tracked in Productivity as a lightweight Worker runtime in the HiClaw ecosystem, covering browser automation, quick tasks, coexistence with OpenClaw and Hermes, and the shared Matrix / MinIO / Higress / Kubernetes deployment stack.
Product updates · Open source
HiClaw is now tracked in Productivity as an Apache-2.0 collaborative multi-agent runtime platform from AgentScope AI, covering Matrix rooms, Manager-Workers orchestration, OpenClaw/QwenPaw/Hermes runtime coexistence, MinIO shared storage, Higress gateway, Kubernetes control plane and Helm deployment.
Product updates · Open source
LongCat-AudioDiT is now tracked in AI Audio as Meituan LongCat's open-weight diffusion TTS and voice-cloning model, covering waveform-latent generation, APG guidance, zero-shot cloning, Seed test results and the released GitHub / PDF sources.
Product updates · Open source
DeerFlow is now tracked in Productivity as ByteDance's open-source long-horizon SuperAgent harness for deep research, coding and creation, covering skills, sub-agents, memory, Docker sandboxing, MCP, IM channels, model routing notes and security hardening notes.
Product updates · Open source
FashionChameleon is now tracked in AI Video as an Alibaba Group and Xiamen University research project for real-time interactive human-garment video customization, covering 23.8 FPS 720p generation claims, KV-cache rescheduling, gradient-reweighted DMD, GitHub, arXiv and HGC-Bench links.
Product updates · Open source
OpenBMB is now tracked as a Chinese open-weight big-model community behind MiniCPM 4/4.1, MiniCPM-V 4.5, BMTrain, BMInf, OpenPrompt, OpenDelta, UltraRAG, ChatDev and AgentCPM-GUI.
Product updates · Open source
AtomCode is now tracked in AI Coding as AtomGit's open-source Rust terminal coding agent and Claude Code alternative, covering multi-model support, DeepSeek/GLM/Qwen/Ollama access, code-graph tools, MCP, image attachments, /undo and AtomGit issue workflows.
Product updates · Market
Kimi WebBridge is now refreshed from the official feature page, adding supported agents, the local install command, click/fill/navigate/extract actions, Chrome DevTools Protocol flow and example workflows such as trending-topic research, job-listing collection and shopping price comparison.
Product updates · Market
Reasonix is now tracked as a DeepSeek-only terminal coding agent, covering prefix-cache stability, DeepSeek-V4-Flash default usage, V4-Pro options, MCP, skills, memory, permissions and desktop prerelease notes.
Product updates · Market
StepFun is now refreshed from platform.stepfun.ai, adding separate coverage for the English Open Platform, Step Plan access plans, StepAudio speech models and Step Image Edit 2.
Product updates · Market
Ant Ling is now tracked as Ant Group's English-facing LLM platform, covering Ling general MoE models, Ring reasoning and coding-agent models, Ming full-modal models, OpenAI/Anthropic-compatible APIs, pricing and agent integrations.
Product updates · Release
Kling AI is now updated from the official kling.ai homepage, covering KlingAI 3.0 Series, Video 3.0 Omni, Native Audio, Creative Studio, API Platform and AI Tools for video, image, sound and effects.
Product updates · Market
ByteDance Seed is now tracked from its English model portfolio, covering Seed2.0, Seedance 2.0, Seedream 5.0 Lite, Seeduplex, Seed3D, UI-TARS, LiveInterpret, Realtime Voice, Seed-Music, Protenix, Seed GR-3 and Seed GR-RL.
Product updates · Market
Baidu ERNIE is now updated from the English ERNIE Blog, covering ERNIE 5.1 agent and reasoning upgrades, ERNIE 5.0's 2.4T unified multimodal model, ERNIE-Image, PaddleOCR-VL and ERNIE 4.5 open-source posts.
Product updates · Market
Kimi WebBridge is now tracked as Kimi's local Chrome and Edge bridge for AI agents, covering local execution, existing-login browser sessions, Kimi Claw Desktop pairing and local agents such as Claude Code, Codex, Cursor and Kimi Code.
Product updates · Market
Kimi Code is now tracked from its official page as a Kimi membership-backed coding agent for Terminal, IDE and Kimi CLI workflows, powered by kimi-for-coding on kimi-k2.6.
Product updates · Market
CodeBuddy is now tracked in the AI Coding category as Tencent Cloud's IDE, plugin and CLI coding assistant, covering Ask/Craft/Plan modes, NES, Figma to Code, MCP, custom agents and cloud deployment integrations.
Product updates · Market
Chinese AI Tools now has a dedicated AI Coding category for AI coding IDEs, coding agents, repository assistants and software engineering automation products.
Product updates · Market
Qoder and TRAE are now tracked as Chinese AI coding products, with Qoder covering IDE, CLI, Quest, Repo Wiki and Teams OpenAPI paths, and TRAE covering AI IDE, Builder Mode, AI Agents and SOLO.
Product updates · Market
Tencent Hunyuan is now filled from Tencent Hy Research, Tencent Cloud English docs, Tencent's English Hunyuan 3D announcement and the Tencent-Hunyuan open-source organization.
Product updates · Release
A separate Hunyuan 3D Global profile now tracks Tencent's global 3D creation engine, text/image/sketch-to-3D inputs, OBJ/GLB output, 20 daily free generations and Tencent Cloud API access.
Product updates · Market
MiMo is now filled from the official English homepage with MiMo-V2.5-Pro, MiMo-V2.5, MiMo-V2.5-TTS, Web Demo, API Access, AI Studio and the current English blog model list.
Product updates · Open source
MiMo-V2-Flash is now tracked as Xiaomi's open-weight fast MoE model with 309B total parameters, 15B active parameters, 256K context, hybrid thinking mode, $0.1/$0.3 per-1M-token reference pricing, MIT weights and SGLang inference support.
Product updates · API
DeepSeek is now tracked from the official English docs with V4-Pro, V4-Flash, 1M context, 384K max output, OpenAI and Anthropic compatibility, thinking mode, tool calls, context caching and current pricing.
Product updates · API
DeepSeek now has a separate agent integration profile covering Claude Code, GitHub Copilot, OpenCode, OpenClaw, Kilo Code and Anthropic-compatible tool configuration.
Product updates · Market
Kimi is now tracked beyond long-context chat across Kimi.com assistant features, Slides, Websites, Sheets, Docs, Kimi Code, Kimi Claw, WebBridge and deep research workflows.
Product updates · API
The Kimi API profile now reflects K2.6, K2.5, K2 and Moonshot V1 model families, 256K K2.6 context, multimodal input, thinking modes, OpenAI-compatible access and API pricing.
Product updates · Market
Qwen is now tracked through Qwen Cloud across Qwen3.6 text and vision, Qwen-Image/Wan image generation, Wan/HappyHorse video, CosyVoice/Qwen ASR audio, embeddings/rerank and Token Plan.
Product updates · API
Qwen Cloud docs now drive the Qwen profile with OpenAI SDK, Responses, Chat Completions, Anthropic Messages, DashScope international endpoints, tool calling, batch, streaming and async tasks.
Product updates · Market
Zhipu AI is now tracked under the Z.ai brand across BigModel/GLM, z.ai, ChatGLM, AutoGLM, AutoClaw, Zread.ai, AMiner, Zlearn, Autotyper and GLM multimodal API lines.
Product updates · API
The BigModel profile now reflects official docs for GLM text, vision, image, video, audio, embedding, rerank, agent, OCR, web search, fine-tuning and OpenAI-compatible access.
Product updates · Market
MiniMax is now tracked across API Platform, Hailuo Video, Agent, Audio/Speech, Music, Image API and Talkie, and the homepage product surface now also highlights M2.7, M2.5, M2-her, M2.1 and the MCP Server card.
Product updates · API
The MiniMax API profile now reflects official docs for text, speech, video, image, music, file management, MCP, local deployment, Token Plan, Credits and pay-as-you-go billing.
Product updates · Open source
Xiaomi MiMo is now tracked for MiMo-V2.5-Pro, 1M-context agent coding, MIT-licensed open-weight models, API pricing, Orbit credits and global model-hub access.
Product updates · API
The integration page now calls out MiMo support through OpenRouter-style routing and custom OpenAI/Anthropic-compatible tool configuration paths.
Product updates · Market
A new 2026 vendor matrix maps China's major model companies across consumer assistants, developer APIs, video, agents, enterprise platforms and international-readiness tiers.
Product updates · Availability
Dola/Cici is now tracked as a ByteDance-linked overseas consumer AI assistant focused on Southeast Asia, Latin America, the Middle East and the UK.
Product updates · API
Tencent Hunyuan is now tracked through Tencent Cloud's international product page, with OpenAI-compatible API access and card-billing paths.
Product updates · Market
Huawei Pangu and iFlytek SparkDesk are now tracked for enterprise, industry, speech, education and healthcare use cases with higher overseas onboarding friction.
News
For global users, QwenChat, Dola and Hailuo cover consumer access; DeepSeek, Hunyuan and Kimi cover developer APIs; Pangu and SparkDesk fit better when enterprise, industry or speech-heavy requirements justify higher onboarding friction.
Product updates · Market
StepFun is now tracked as a Shanghai multimodal model company behind Step 3.5 Flash, StepClaw, open-model audio/video/image releases and an AI plus device strategy.
Product updates · Open source
Step 3.5 Flash is tracked as a sparse-MoE model optimized for agent workloads, with Apache-2.0 open-source licensing and visible OpenRouter/OpenClaw integration paths.
Product updates · Availability
The overseas support report identifies chat.qwen.ai as Qwen's international web entry, with email login, broad language support and cloud billing paths to verify separately.
Product updates · Availability
Hailuo AI is tracked as MiniMax's international consumer product entry, with credits and subscription-style billing called out for overseas users.
Product updates · API
Moonshot's international API path is treated separately from the China API path, making Kimi more API-first for overseas users.
Product updates · Pricing
The overseas support report flags sharp domestic-versus-overseas subscription differences for GLM Coding Plan, so buyers should verify regional pricing before committing.
Product updates · API
DeepSeek's platform entry is now listed as the API route to verify for global users, alongside open-source deployment paths.
Product updates · API
The overseas report treats Baidu's consumer web experience as China-first, with Qianfan API access as the overseas route.
Product updates · Market
A new guide summarizes international websites, overseas payment, API access, open-source routes and restrictions across major Chinese AI products.
News
For overseas users, QwenChat, Dola, Hailuo AI and Tencent Hunyuan provide relatively direct international access; DeepSeek and Kimi are developer API checks; ERNIE, Pangu, SparkDesk and GLM are more China-first or enterprise-first paths.
News
DeepSeek and Qwen cover the main global developer baseline, GLM covers coding-agent evaluation, Kimi covers very long context and ERNIE/Qianfan cover Baidu Cloud or China-local enterprise deployment.
News
China-origin AI matters when cost, open-source access or China-market deployment matter, but regional availability, compliance and vendor-risk checks still apply.
News
API access is the fastest starting point. Local deployment depends on license and hardware constraints, while hybrid architecture fits sensitive data or latency-sensitive workloads.
News
Start with Kling if you need a relatively clear global Chinese video workflow, then use Hailuo as a second test for creative range and friction.
News
Start with DeepSeek, Qwen and MiMo for broad developer baselines, then add Kimi, GLM, Ant Ling, MiniMax, ERNIE/Qianfan, Hunyuan and Doubao/Ark based on ecosystem, language, region and billing needs.
News
Use it as the release anchor for the dense 27B branch, then verify the current Qwen pages for access and model-availability details.
Product updates · Release
Qwen's hosted proprietary line now includes Qwen3.6-Max-Preview, with Alibaba Cloud Model Studio access and stronger world knowledge, instruction following and agentic coding than Qwen3.6-Plus.
News
Use this as the official Qwen3.6-Max-Preview release reference, and keep Qwen3.7 unconfirmed unless Alibaba names it explicitly.
News
Use this as the release anchor for the 35B A3B branch, then follow the current Qwen pages for access, pricing and deployment terms.
News
Use this as the release anchor for Qwen3, then point readers to the current model pages for pricing, availability and newer variants.
News
Use it as the Qwen news anchor for reasoning-model coverage, then follow the current Qwen pages for updated product names and access paths.
News
Use it as a release marker for Hunyuan Turbo S, then check the current Hunyuan pages for current access and pricing paths.
News
Treat this as historical pricing context, not current pricing proof.
News
Use it to understand platform reach, then verify the current product page for exact availability and integration terms.
News
Use it as release history for MiniMax-01, then cross-check the current API page for updated model names and pricing.
News
Use this as a news reference for the consumer side of DeepSeek, then cross-check the current app and API pages for pricing and access details.
News
Use it as a release-history reference for Kling, then verify the current product page for the latest feature set.
News
Use it as pricing history, then verify the current pricing page before publishing any current numbers.
News
Use it as a long-context milestone, then verify the current Kimi page for today's limits and access model.
News
Use it as a demand-and-reliability reference, not as current availability proof.