Kimi K3.1 Pro scores 55 on Artificial Analysis at $2 in and $10 out, and Moonshot promises open weights by November 4

A soon to be open weight model now sits three points behind the top of the Artificial Analysis index, at half the price of the leader.

What happened

Moonshot AI released Kimi K3.1 Pro on its API. It scores 55 on the Artificial Analysis Intelligence Index, an 11 point jump over Kimi K3, behind only Claude Opus 5.5 at 58 and ahead of several models at 53, according to The Decoder and Simon Willison. Moonshot says full model weights will be released by November 4, 2026.

Pricing is $2 per million input tokens and $10 per million output tokens, down from $3 and $15, with cache hits at $0.20. Simon Willison notes that is the same as GPT-6.1 Sol and half the price of Opus 5.5. A Highspeed version runs 2 to 3 times faster at $4 and $20. The model has a one million token context window and handles text and images.

Moonshot reports 65.7% on Terminal-Bench 4.0 against 66.4% for Opus 5.5. On its internal hallucination stress test, claim level hallucinations fell from 23.4% to 8.3%, though GPT-6 Astra scores 5.5% on the same test. Artificial Analysis puts the cost per index task at $1.98, close to K3, because K3.1 Pro writes about 14% more output tokens.

My take

Two things matter for client work. First, the headline price is not the bill. A model that writes more tokens can cost the same per task as an older one, so compare cost per finished job on your own workflow, not price per million.

Second, open weights change the hosting question. If the weights ship as promised, a client with strict data rules could run a near frontier model on infrastructure they choose. That is worth a pilot, but check each client's policy on model origin before you switch anything, because some organizations restrict which providers they allow.

My test plan: take 50 real inputs from one live workflow, run them through your current model and K3.1 Pro, and score accuracy, cost per run and latency. Keep the model name in config so switching back is one change.

This is the kind of model evaluation I run before changing anything in a client's automations. More at romielwillautomate.dev.