Mistral Large 4 price cut: the flagship's API price halves a week after its open weights shipped.
The Mistral Large 4 price cut landed on October 30, 2026: Mistral cut its flagship from $1.36 input / $4.18 output per million tokens to $0.70 / $2.10 — back to roughly the launch-promotion level — in the same week the model's open weights shipped. Here is the full price history from our own changelog, and what the cut is worth on a real workload.
By AI Pro League · Published
Mistral Large 4 price cut is the second price move in this model’s short life, and our AI lab release tracker changelog tells the whole story. Mistral Large 4 was released on Oct 6, 2026 at a launch-promotion $0.68 input / $2.09 output per million tokens. When the promotion ended on October 20, 2026, the price rose to $1.36 / $4.18 (cached input $0.14). Nine days at that rate later, Mistral’s official announcement of October 29 — recorded in our changelog and live in our database on October 30 — cut it to $0.70 / $2.10 with cached input at $0.070: a 48.5% input cut, a 49.8% output cut and a 50% cached cut. The current rates are also on Mistral’s official pricing documentation, which our record cites.
What the Mistral Large 4 price cut changes — and what it doesn’t
- The bill roughly halves. Every rate — input, output and cached input — came down by about half in one move (worked example below).
- The model did not change. Same 1M-token context, same text-and-image inputs, same recorded benchmarks — this is a price event, not a new release.
- The price is back where it started. $0.70 / $2.10 sits within a few cents of the $0.68 / $2.09 launch promotion — the post-promotion rise lasted nine days.
- You can now also run it yourself. The open weights shipped October 29 on Hugging Face under the Mistral Ecosystem Open-Weight License (a modified MIT; Model-as-a-Service businesses need a separate commercial licence, per Mistral’s announcement) — the API cut arrived alongside a self-hosting option.
The price history, from our changelog
| Period | Input $/1M | Output $/1M | Cached $/1M |
|---|---|---|---|
| Launch promotion (to Oct 20, 2026) | $0.68 | $2.09 | $0.14 |
| Post-promotion (Oct 21–29, 2026) | $1.36 | $4.18 | $0.14 |
| Current (from Oct 30, 2026) | $0.70 | $2.10 | $0.070 |
One honest caveat about the model behind the price: in our AI model rankings, Mistral Large 4 carries a League Score of 11.1 (#41 overall, #28 on value) — but that score rests on a single recorded component, LiveBench overall at 43.9 measured Oct 26, 2026, so its score coverage is 40% of the five-benchmark set. A thin record is a coverage statement, not a verdict — the same rule our LiveBench explainer sets out. The price, by contrast, is fully documented at every step.
What the cut is worth — the worked example
Price the cut on a concrete workload — 10M input tokens and 2M output tokens, billed at the old and new rates in our AI model pricing index:
| Rate card | Input cost | Output cost | Total |
|---|---|---|---|
| Post-promotion ($1.36 / $4.18) | $13.60 | $8.36 | $21.96 |
| Current ($0.70 / $2.10) | $7.00 | $4.20 | $11.20 |
The same workload falls from $21.96 to $11.20 — a saving of $10.76, just under half. Repeated-context workloads save on the cached rate too: 10M cached input tokens cost $1.40 at the old cached rate and $0.70 now — see prompt caching pricing explained for how cached rates compare across the field.
API or self-host? The cut sharpens the question
Mistral Large 4 is one of 30 open-weights models among the 59 we track, and since October 29 its weights are downloadable. The API is listed with 5 providers in our record (Mistral AI Studio, OpenRouter, Hugging Face, Microsoft Foundry, NVIDIA). At $0.70 / $2.10, the hosted route just got about half as expensive, which moves the self-hosting break-even point outward: the weights now compete on control, privacy and the licence terms above — not on a two-to-one price gap that no longer exists. Our monthly AI price drops ledger records this cut alongside every other tracked change.