Moonshot launches Kimi K3.1 Pro, scoring near the closed frontier ahead of a promised Nov 4 open-weight release
2026-10-23
Moonshot AI released Kimi K3.1 Pro via API on October 23, 2026 — a 2.8 trillion total parameter / 104B active parameter mixture-of-experts model with a 1M-token context window, post-trained from Kimi K3 with substantially larger post-training compute. Moonshot says full weights will follow by November 4, 2026. API pricing is $2/M input and $10/M output tokens; a 'Highspeed' variant runs 2-3x faster at roughly double the price. Artificial Analysis's Intelligence Index scored it 55 — fourth place, behind Claude Opus 5.5 (58) and tied with Claude Sonnet 5.5 and Meta's Muse Wonder Full (56), ahead of Claude Fable 5.1, GPT-6 Astra, and Gemini 4 Argon (all 53). It scored 64% on Artificial Analysis's Cyber Index, the highest publicly accessible score (second overall behind GPT-6 Sol Daybreak Blue's 69%), and ranked #8 on LMArena's Code Arena: WebDev board at 1,692 points, up from Kimi K3. Social-media reaction, per Latent Space's roundup, raised concerns about inference economics, weaker results on some interactive tasks, and allegations of undisclosed paid promotion.
Significance 4: A new model lands within 3 points of the closed frontier leader on an independent benchmark (Artificial Analysis) and is promised as open weights within two weeks — the closest a soon-to-be-open model has come to frontier parity on this index per Simon Willison — but weights aren't released yet and some reaction alleges undisclosed promotion, so this is held at 4 rather than 5.
Implications · machine-drafted, not owner judgment
If the November 4 weight release holds and independent scores survive broader testing, this narrows the open/closed gap further and adds competitive pricing pressure on closed frontier labs, extending a trend already visible in GLM-5.3 and DeepSeek releases this year. For the owner, it's another data point that the open tier is converging toward frontier capability faster than the book's substitution-trajectory argument assumes computational scarcity will allow — worth a side-note, though the model is far too large for any local/64GB deployment relevant to the owner's own verdicts.
- Independent benchmark re-confirmation once weights are actually released on November 4
- Whether the undisclosed-paid-promotion allegations are substantiated
- Any price or positioning response from Anthropic, OpenAI, or Google
Sources
- curator [AINews] Kimi 3.1 Pro 104BA-2.8T: Fable-5.1-Class Open Weights narrows Closed-AI lead to 1 month retrieved 2026-10-24
- curator Introducing Kimi K3.1 Pro retrieved 2026-10-24
Affected axes
Hard reasoning · Coding · Agentic / tool use