WEEK IN REVIEW · OCTOBER 2026

Week in Review: October 2026 was a price and access month, not just a model month

Three patterns explain almost every October launch in this archive. If you only remember the model names, you will miss the part that affects your budget and your product.

1. The strongest model is not always the one you can buy

Meta’s Muse Wonder Full is the clearest case: the stronger model exists, but at launch it is available only to Muse Code subscribers, with no pay-as-you-go API price. Mistral Large 4 is the second case: a public preview API, with weights promised later. Kimi K3.1 Pro is the third: available as a service now, weights by 4 November. A launch announcement is not an access guarantee — read the access line on every article before you plan around a model.

2. Introductory prices are not permanent prices

Gemini 4 Argon doubles on 1 December 2026, from $2/$10 to $4/$20 per million tokens. Mistral Large 4 preview rates are half its posted rates. Budget on the post-introductory figure, and treat the introductory figure as a trial window, not a cost base.

3. Small, cheap models are the real workhorses

Claude Haiku 5.5 at $0.10/$0.50 per million tokens under 100,000 tokens, GPT-6 Luna at the same headline rate, and MiniMax M3.1 Flash are built for high-volume work — summaries, classifications, subagent tasks — where a flagship would be wasteful. The smart October decision for most teams is not “which flagship is best”, it is “which cheap model handles 90% of the volume, and when do we escalate to a flagship”.

What this site will not do

This site will not turn a waitlist into a launch, a preview price into a permanent price, or a maker’s benchmark into an independent test. Where a figure is reported, it is labelled reported. Where a field was not published, it is marked not published. That is the whole trust model — see How we verify.