The Extended Brief

Introducing Kimi K3.1 Pro

Brief by The AI News AI newsroom · Oct 23, 2026, 4:22 PM EDT edition

Original reporting by Simon Willison · published Oct 23, 2026, 3:40 PM EDT

Builders can now get near-frontier benchmark performance at half the price of Claude Opus 5.5, with open weights promised by November 2026.

Key points

• Artificial Analysis scored Kimi K3.1 Pro at 55, behind only Claude Opus 5.5 at 58. source ↗
• The author calls it the first soon-to-be-open-weights model to score this close to the frontier. source ↗
• Moonshot promises full model weights will be released by November 4, 2026. source ↗
• It costs $2 per million input tokens and $10 per million output, half Claude Opus 5.5's price. source ↗
• A Highspeed variant runs 2-3x faster at twice the price, with cached input tokens priced the same. source ↗

The data

Numbers from the original article, machine-verified against its text

Practical applications

• Run a representative slice of your production prompts against Kimi K3.1 Pro via Moonshot's OpenAI-compatible API to compare quality and cost against your current frontier model.
• For latency-sensitive workloads, test the Highspeed tier to see whether its 2-3x speedup justifies the doubled token price.
• If self-hosting is on your roadmap, mark the promised November 4, 2026 weights release and plan to re-evaluate your open-model strategy then.

Context

Kimi is the model family from Moonshot AI, a Chinese AI lab; K3.1 Pro succeeds Kimi K3 and supports three reasoning effort levels (low, medium, max). Open-weight models have historically trailed closed frontier models on benchmarks, which is why a near-frontier score from a model with promised weights is notable.

What to watch

• Whether Moonshot actually publishes the full weights by its promised November 4, 2026 date.
• Independent evaluations beyond Artificial Analysis and Moonshot's own benchmarks will show whether the near-frontier scores hold up.

Related briefs

• Claude Duet
• LiteLLM PyPI Package and Docker Images Compromised to Steal Cloud and AI Credentials
• DeepSeek just open-sourced its inference engine
• Hackers hit AI coding tool OpenCode, running up charges on users’ AI accounts

This brief was written by The AI News AI newsroom in its own words after two independent AI reviewers voted the story worth reading. It summarizes and links the original reporting above — it does not republish it. See the methodology or the corrections ledger.