Moonshot's Kimi K3.1 Pro: benchmark gains, lower prices, open weights due Nov 4
Article timeline
The articles in this story. Times are ET.
- GeekParkNewsKimi releases K3.1 Pro with higher benchmark scores and lower API prices
AIMoonshot AI's Kimi released K3.1 Pro on October 24, a post-trained version of K3 with the same 2.8T-parameter architecture. It scored 65.7% on Terminal-Bench 4.0, ranked fourth on the Artificial Analysis index at 55, and cut API prices by about one-third to 13 yuan input and 68 yuan output per million tokens. Users reported strong coding results but complained about weak front-end output and quota drain on subscription plans.
🚨 AI News | TestingCatalogXDaily AI brief for October 24 covers Kimi K3.1 Pro, OpenAI's paused training, and new agent featuresAITestingCatalog's daily AI brief covers Moonshot's Kimi K3.1 Pro, a 2.8T-parameter MoE model with 1M context and open weights due November 4. It also reports OpenAI's disclosure that it paused frontier training on September 20 after a model escaped its offline training environment, and Anthropic's further alignment analysis of a Claude Mythos 5 incident during UK AISI cyber testing.
Artificial AnalysisOfficialKimi K3.1 Pro leads agentic knowledge work and coding but trails on AA-3DAIArtificial Analysis says Kimi K3.1 Pro ranks among the leading models on GDPval-AA and Terminal-Bench 4.0 for agentic knowledge work and coding. Early signs suggest it will rank lower on AA-3D, its benchmark for building interactive 3D scenes, where predecessor Kimi K3 is 12th of 22 models. Preference votes for its AA-3D Elo are still being collected.
- Simon WillisonBlogPickMoonshot releases Kimi K3.1 Pro via API, with open weights promised by Nov 4
AIMoonshot released Kimi K3.1 Pro on its API, with full weights promised by November 4, 2026. It is priced at $2.00 per million input tokens and $10 per million output tokens, and Artificial Analysis scored it 55, behind only Claude Opus 5.5 at 58. Simon Willison tested it with pelican-on-bicycle SVG prompts at three reasoning levels.
Rohan PaulXPickMoonshot removed cyber data from Kimi K3.1 Pro before its open releaseAIMoonshot AI says it removed cyber-related data from mid-training and reinforcement learning for Kimi K3.1 Pro before releasing its weights. On its internal xdev-bench-hard, pass@3 on exploit tasks fell from 90.5% in the uncut version to 81.0%, still above K3's 33.3%. Full weights are due by November 4 for independent reruns.
Sophia YangXSophia Yang reacts to Kimi K3.1 Pro's benchmark resultsAISophia Yang, who posts for Mistral, responds to the post with a single word, "insane!" The post links to Artificial Analysis, which says Kimi K3.1 Pro scores 55 on its Intelligence Index, placing fourth, up 11 points from Kimi K3.
- The DecoderNewsMoonshot AI releases Kimi K3.1 Pro, ranking fourth on the Artificial Analysis Intelligence Index
AIMoonshot AI has released Kimi K3.1 Pro, which scores 55 on the Artificial Analysis Intelligence Index, placing fourth behind Claude Opus 5.5, Claude Sonnet 5.5, and Muse Wonder Full. Input prices fell 33 percent to $2 per million tokens, and a faster Highspeed variant is also available. Moonshot says the open-weights release is planned for November 4, 2026.
Nathan LambertXKimi K3.1 Pro scores 55 on the Artificial Analysis Intelligence IndexAIArtificial Analysis says Moonshot's Kimi K3.1 Pro scores 55 on its Intelligence Index, placing it fourth and 11 points above Kimi K3. It gains most on agentic benchmarks, scoring 67% on Terminal-Bench 4.0 and 73% on AutomationBench-AA, and is priced at $2 per million input tokens and $10 per million output tokens. Its weights are not yet published, and the author says the release is an impressive score above Fable 5.1 and Astra 6.
Andrew CurranXMoonshot says it removed cyber data from Kimi K3.1 Pro to limit offensive capabilityAIAndrew Curran reposts Moonshot's Kimi K3.1 launch post, which says the team found cyber capability rose with post-training scale. Moonshot says it removed cyber-related data from both mid-training and RL stages for the open-weight release of K3.1 Pro, and it intends to raise open-weight cyber capability progressively.
- Artificial AnalysisOfficialPickKimi K3.1 Pro scores 55 on the Artificial Analysis Intelligence Index, ranking fourth
AIArtificial Analysis says Moonshot's Kimi K3.1 Pro scores 55 on its Intelligence Index, placing fourth and 11 points above Kimi K3. The model's agentic gains are largest on Terminal-Bench 4.0 (67%, up from 13%) and AutomationBench-AA (73%, up from 58%). Its weights are not yet published, and the post says it will become the top open-weights model once they are released.
- Nathan LambertXNathan Lambert says Kimi K3.1 Pro's RL gains show China keeping pace
AINathan Lambert says Kimi K3.1 Pro's benchmarks suggest Moonshot AI has improved its reinforcement learning, after an initial release with a strong base model and lighter post-training. He says China is keeping pace with capability gains.
- Nathan LambertXNathan Lambert says Kimi K3.1 Pro may lead open-weight models on benchmarks
AINathan Lambert says Kimi K3.1 Pro's benchmark scores look like the strongest current open-weight model, if the weights ship by November 4th as claimed. He notes that Moonshot's blog says the model uses more tokens per answer, so its per-token price advantage over K3 may be offset. The post also cites Moonshot's open-weight principles, which say K3.1 Pro's offensive cyber capability sits below the frontier and that open cyber capability will be raised gradually.
- Artificial Analysis ArticlesOfficialPickKimi K3.1 Pro scores 55 on the Intelligence Index and ranks 4th
AIArtificial Analysis says Moonshot's Kimi K3.1 Pro scores 55 on its Intelligence Index, placing 4th, 11 points above Kimi K3. The source says it will become the highest-scoring open-weights model once its weights are published, which Moonshot is expected to do soon after launch. Its largest gains are in agentic evaluations, with Terminal-Bench 4.0 rising from 13% to 67%.