2026-09-22 · TuesdayPublished daily at 08:00
AI Daily
- 1Xiaomi releases and open-sources the MiMo-V2.6 series, scaling up reinforcement learning toward self-improvementWeChat Official Account: Xiaomi MiMo
- 2
- 3
01Model Releases/Updates
Xiaomi releases and open-sources the MiMo-V2.6 series, scaling up reinforcement learning toward self-improvement
Xiaomi MiMo officially released and open-sourced the MiMo-V2.6 series today, including two natively omnimodal models, Pro and Flash. It calls this a key step in exploring RSI (recursive self-improvement).
Original textxAI Releases Grok 4.7, Focused on Coding and Knowledge Work
xAI released Grok 4.7, positioning it as its most powerful model for coding and knowledge work. It is priced at $2 per million input tokens and $6 per million output tokens, with the same price and speed as Grok 4.6. A fast variant with double the speed and price is also available.
Original textXiaomi releases MiMo-V2.6 Pro and Flash, open-source omnimodal models
Xiaomi MiMo releases two omnimodal models, MiMo-V2.6 Pro and Flash, trained through reinforcement learning at scale. Pro performs on par with Claude Opus 5 and GPT-5.6 Sol on most agent benchmarks and scores 46 on the Artificial Analysis Intelligence Index, the highest among open-source models. Its capabilities span coding, computer use, 3D reasoning, and creative work.
Original textXiaomi releases open-weight model MiMo-V2.6-Pro, topping the Artificial Analysis Intelligence Index for open-weight models
Xiaomi releases the open-weight model MiMo-V2.6-Pro, scoring 46 on the Artificial Analysis Intelligence Index, up from 26 for its predecessor MiMo-V2.5-Pro—the highest score among open-weight models.
Original textQwen Releases Qwen-Image-2.1: A 7B Single Checkpoint Supporting Both Image Generation and Editing
Tongyi Qianwen releases Qwen-Image-2.1, a 7B-parameter native image generation and editing model. Generation and editing share a single checkpoint, with support for up to 10 reference images. The model includes a built-in prompt-enhancement LLM, has been integrated with diffusers and ComfyUI, and provides an installation-free browser demo on Hugging Face Spaces: https://huggingface.co/spaces/hugging-apps/qwen-image-2-1
Original textXiaomi MiMo-V2.6-Pro ranks around 10th on the Code Arena: WebDev leaderboard and around 3rd among open-weight models
Xiaomi releases two omnimodal models, MiMo-V2.6-Pro and Flash. MiMo-V2.6-Pro ranks approximately 10th on the Code Arena: WebDev leaderboard with 1628 points (AutoEval), and approximately 3rd among models with open-source weights (MIT license), up 153 points from MiMo-V2.5-Pro's 1475.
Original text02Product Releases/Updates
Kimi releases Kimi Code Desktop 1.0, launching simultaneously on macOS and Windows
Kimi (Moonshot AI) has released Kimi Code Desktop 1.0, the official Kimi Code desktop client. macOS (Apple and Intel chips) and Windows versions are launching simultaneously. Download it at kimi.com/code.
Original text03Industry News
British Columbia sues OpenAI, alleging it failed to refer flagged ChatGPT activity to police before a shooting
British Columbia sued OpenAI in California, alleging that flagged ChatGPT activity should have been reported to police before the February 10, 2026, Tumbler Ridge shooting, which left 8 people dead, including 5 children and an educator, and 27 injured.
Original textAmazon bans the Meta Muse agent from shopping on behalf of users, escalating the dispute between the two sides
Amazon has blocked access for Meta's personal AI agent Muse to shop on its website on users' behalf, saying it never authorized the access and that Muse does not identify itself and collects and stores users' account credentials, posing privacy and security risks.
Original text04Tips and perspectives
Musk says Grok 4.7 puts xAI in third place in agentic coding
Elon Musk cited an Artificial Analysis evaluation saying that Grok 4.7 ranked third for agentic coding, behind Anthropic and OpenAI, making xAI the third-ranked company.
Original textNathan Lambert writes an analysis of the balance of power between China and the United States in open-source models
Nathan Lambert published a review of the state of open-source models for his briefing to the U.S. Congress, noting that Chinese open-weight models now lead in downloads, benchmark performance, and academic adoption. Since July 2025, they have led in Hugging Face downloads by approximately 1.6B (3.2B total, twice the U.S. figure).
Original textArtificial Analysis evaluates Grok 4.7: agentic knowledge work reaches the frontier, coding agent score rises to 56
Artificial Analysis released its evaluation of Grok 4.7. The model scored 46 on the Intelligence Index, 2 points higher than Grok 4.6, and achieved an Elo rating of 1657 (+111) on AA-Briefcase, behind only Claude Opus 5 and Claude Fable 5.1.
Original textDigital Life Kazik's interview with the Chinese localization team: The real situation of subtitle and manga groups in the AI era
The author interviewed members of several subtitle and manga localization groups and found that they are not opposed to AI. After Eliza's subtitle group handed transcription and timing over to AI, timing work that originally took 3 to 5 hours was reduced to 20 minutes to 1 hour. The programmer Zhenyu developed an AI tool for a favorite streamer and completed Chinese subtitles for a two-hour livestream alone. Manga localization groups still insist on manual production because of issues such as typesetting quality; members value the community of fellow enthusiasts and the hobby itself.
Original textTomer Tunguz on AI-optimized if-then decisions: specialized decision models cut classification costs nearly 100-fold
Tomer Tunguz wrote that the latest wave of AI is taking over the if-then decision primitives in software. Specialized decision-makers such as Jev and SemIf return results in hundreds of milliseconds, at costs approximately 76x to 209x lower than traditional generative calls. The author replaced about one-quarter of the calls in his own Agent. On 98 manually verified production email threads, Jev achieved 80% classification accuracy and local SemIf achieved 82%, both higher than the production model's 47%.
Original textLinear reworks its CI process to address the verification bottleneck caused by AI coding, reducing PR wait times from 6 minutes to 5 minutes
A Linear engineer shared how the company addressed CI becoming a bottleneck after AI accelerated code writing. PR wait times fell from more than 6 minutes to just over 5 minutes, while unit-test runner time was roughly cut in half.
Original text