OpenAI launched GPT-5.1, delivering marked gains in complex reasoning, code generation, and image understanding, while further reducing API latency and cost and improving enterprise readiness.
Model Watch
Weekly LLM rankings, capability comparison and benchmarks. Track GPT, Claude, Gemini and domestic model updates.
Anthropic released the Claude 4 series with very large context windows and stronger agent capabilities, standing out in code refactoring and long-document analysis.
Google launched Gemini 3, strengthening native multimodal comprehension across image, audio, and text, plus long-video understanding, and opening finer-grained video analysis APIs to developers.
Alibaba open-sourced the Qwen3 series spanning multiple parameter scales, with strong Chinese and coding ability, plus tool-calling and a switchable reasoning mode.
DeepSeek released V3.1, delivering near-frontier overall capability at extremely low inference cost, continuing to push the boundary of freely usable open AI models.
Well-known AI tech blogger Kazik (@Khazix0918) shared on X sharp observations on China’s AI products and model developments, focusing on big-tech launches, open-source progress, and industry trends.
The Hugging Face team blog introduces Mistral Large 4, a 1-trillion-parameter native multimodal MoE with MLA+DSA sparse attention and DSpark speculative decoding; vLLM and Transformers shipped Day 0 support.
Alphabet CEO Pichai announced Gemini 4 Carbon, highlighting its lead in cybersecurity and long-horizon agentic work, topping the Artificial Analysis Intelligence Index v5.0.
DeepMind formally released Gemini 4 Carbon, built for the full cybersecurity lifecycle and long-horizon agentic work, topping multiple external benchmarks with 100% prompt-injection resistance.
OpenMOSS/Fudan released MOSS-Image, a 5B-parameter open-weight single-stream DiT for image generation with Chinese/English long-text rendering and editable JSON layout, reaching the top open-weight score on long-text benchmarks like LLTB.
ModelBest released and open-sourced MiniCPM-OCR: a 1.3B parser topping dual SOTAs and a 2B Jev model that judges whether parse results are trustworthy.
Google unveiled its next-generation world model Genie 4, a clear step up from Genie 3 in photorealistic quality and physics simulation. Users steer characters or cameras live with WASD and dual sticks, and can rewrite the scene in natural language mid-run, while the model also generates ambient sound and real-time character dialogue natively.
Google DeepMind released Genie 4, a world model emphasizing real-time interactive simulation and stronger photorealism, and for the first time natively generating audio inside a live world model, including ambient sound and interactive character dialogue.
On October 27, OpenAI announced AI model breakthroughs in mathematics, computer science, and physics, including a proof of the local smoothing conjecture in all dimensions.
OpenAI announces GPT-6.1 Terra's return, calling it the first model noticeably better at writing, personality, and nuanced request understanding, with 2x the usage allowance of 6.1 Sol at lower price.
Liquid AI released the LFM3 multimodal model family, emphasizing specialized intelligence and on-device deployment.
OpenAI releases GPT-6.1 Terra, a fast and efficient computer-use and visual-reasoning model priced 50% below GPT-5.6 Terra and GPT-6.1 Sol at $1/$5 per million tokens.
Artificial Analysis rates OpenAI's GPT-6.1 Terra as reaching the intelligence-cost Pareto frontier at low-to-mid effort, at half or less the price of GPT-5.6 Terra.
Moonshot released its flagship Kimi K3.1 Pro with 2.8T parameters and a 1M-token context; full model weights are expected before November 4. API pricing is ¥13 per million input tokens and ¥68 per million output tokens — roughly half of K3 — alongside a Highspeed tier at 2–3× speed and 2× price.
Kimi K3.1 Pro, the first model in Moonshot's K3.1 series, is a ~3T-class open-weight model with a 1M-token context; full weights drop by Nov 4 and it scores 65.7% on Terminal-Bench 4.0, ahead of GPT-6 Astra.
MiniMax has open-sourced M3.1 Flash weights (428B total, 23B active per token) on Hugging Face and ModelScope.
Moonshot launched Kimi K3.1 Pro, scoring 55 on the Artificial Analysis Intelligence Index for 4th place — an 11-point gain over Kimi K3 — trailing only Claude Opus 5.5, Claude Sonnet 5.5, and Muse Wonder Full.
fal launched FLUX 3 Max with Black Forest Labs; it ties Seedance 2.5 at 1139 Elo in independent human preference tests, generates 10s/720p video in ~12s, and is 75% cheaper for three weeks.
Cohere released Rerank 5 in Pro (accuracy) and Fast (speed) tiers for enterprise retrieval and RAG reranking.
Arena says DeepSeek-V4.1-Flash-20261017 (Max) is the #1 open-source Agent Arena model (11th overall): +5.1% net over 7.5K real chats, $0.20 median cost (~27% of GPT-6.1 Sol), +11.5% Confirmed Success, no tool hallucination.
DeepSeek released V4.1 Flash 1017, scoring 45 on the Artificial Analysis Intelligence Index (up 6), tying GLM-5.3 as the #2 open-weight model behind MiMo-V2.6-Pro.
Arena says Meta's Muse Wonder Full (Max) debuted on the Code Arena: WebDev board at #6 with 1726 points, making Meta the 3rd lab there and taking the spot Google held.
OpenAI says its latest models show major gains on low-resource languages like COBOL, Fortran, and Visual Basic, and introduced two in-house evals to measure the improvement — filling a legacy-code gap in AI coding.
fal launched FLUX 3 Max with @bfl_ai, a post-trained video model tying for #1 in third-party image-to-video quality, generating 10s/720p video in ~12s, with a 75% discount for three weeks as the cheapest frontier video model at its tier.
Arena announced DeepSeek-V4.1-Flash-20261017 (Max) as the number-one publicly available and free-to-use model on the Code Arena: WebDev leaderboard, surpassing Kimi K3 (Max) at roughly one-tenth the cost.
Arena announced that Meta's Muse Spark Open (Max) ranks first among US open, free-to-use code models on Code Arena: WebDev with a score of 1511, and 13th overall among open free-to-use code models.
DeepSeek upgraded V4.1 Flash to DeepSeek-V4.1-Flash-1017; the web and app updated in sync, and API access method and pricing are unchanged, still reached via deepseek-flash.
Kuaishou's Kling AI officially launched video model Kling 4.0; a partial Flash trial was opened to some black-gold and Ultra annual users in late September. The reviewer found gains in three areas: up to 30-second one-take output, better physics detail in complex scenes, and improved motion continuity plus up to 10 multi-frame references.
Arena announced Meta’s Muse Spark Open has entered the agent arena where users can test it live and vote, with official scores to follow. The model has 816B total and 42B activated parameters, a 1M-token context, and native multimodal support for image, audio, and documents.
Meta released and open-weighted Muse Spark Open, available on Hugging Face under the Muse Model License. It has 816B total and 42B activated parameters, a 1M-token context, targets coding and agent workflows, natively supports image, audio, and document multimodality, and adds safety post-training versus the August version.
ARC Prize published Gemini 4 Argon results on ARC-AGI (Verified): 97.5% on ARC-AGI-3 via provider adapter harness (cost 4.8K USD), 49.8% on the standard harness; 91.3% on ARC-AGI-2 at 0.32 USD per task; 98.5% on ARC-AGI-1 at 0.16 USD. DeepMind is the second frontier lab past 95% on ARC-AGI-3.
Microsoft AI announced completion of MAI-Base-2 base training, an 80B-activated, 2.8T total MoE base model with lower BPB than all tested open-weight models on held-out evals, and 4.2 to 12.3x higher base-training FLOPs efficiency than DeepSeek-V4-Pro-Base.
StepFun reposted FeatherlessAI news that Step 5 Preview is live there, served at FP8 precision. The open-weight model is a 600B sparse MoE with 27B activated per token, aimed at long-horizon agents, frontier-level on software engineering and expert knowledge work (especially finance); per Artificial Analysis its cost per task is about 1/6 of Claude Opus 5.5, with intelligence on par with Kimi K3.
Arena announced that Meta's Muse Wonder series models are now live on the agent-running Arena where users can test and vote, with leaderboard scores to follow.
Meta released Muse Wonder and Muse Wonder Full. On the Artificial Analysis Intelligence Index v4.3 they score 52 and 56 (max reasoning effort), a marked improvement over the previous generation per real-world testing.
Google DeepMind released EmbeddingGemma 2, built on the Gemma 4 architecture and freely usable under the Apache 2.0 license. It maps text, code, images, video, and audio into a single unified embedding space.
ARC Prize published DeepSeek V4.1 Flash results on ARC-AGI (Verified): 72.9% on ARC-AGI-2 at $0.13 per task, and 94.5% on ARC-AGI-1 at $0.07 per task. Versus V4 Flash's best, ARC-AGI-1 is +5.5 points and ARC-AGI-2 is +11.5 points, but per-task cost is about 250% higher.
Arena announced Mistral's Mistral Large 4 is now in the Agent Arena, where users can test and vote, with scores to follow. The model is a 1T-parameter, 49B-active native multimodal open-weights model, also available in Code Arena's WebDev, Text, and Vision tracks.
Google released Gemini Nano Banana 2.1 (gemini-nano-banana-2.1), positioned as an efficient image-generation and conversational-editing model and an update to Nano Banana 2 (gemini-3.1-flash-image). The older image model will be deprecated on October 29, 2026.
Mistral released Mistral Large 4 (Research Public Preview). On the Artificial Analysis Intelligence Index it scores 38, on par with GPT-6 Luna (max, 38), and is called the smartest model outside the US and China. Open 1T-parameter (49B active) weights are planned by end of October.
Liquid AI released the d1 decision model, adding text and image inputs to its prior capabilities, accessible via console.liquid.ai and the d1 Playground. The model targets tasks that require seeing, reasoning and deciding together.
Arena reports OpenAI's GPT-6.1 Sol (Max) ranks 5th on Agent Arena with a +11.23% net gain and a median task cost of $0.56, reshaping the Pareto frontier.
Arena's latest Agent Arena ranks Anthropic's Claude Sonnet 5.5 third with a +12.5% net gain, but at $2.74 median cost per task it is more expensive than Opus 5.5 and stays off the Pareto frontier.
Ai2 open-sourced AstaBrief 8B, built on Qwen3-8B, that turns research questions and retrieved snippets into cited reports; it is live as Fast mode in Asta's Generate a report feature.
The LMSYS team fine-tuned LLaMA on multi-turn ShareGPT conversations and trained Vicuna-13B at minimal cost; it approached ChatGPT quality across several benchmarks.