Mistral
Mistral AIMistral Large 4 launch pricing and Mistral Medium 3.5 on Foundry.
Mistral Large 4 entered public preview on October 6, 2026. Mistral's October 6 changelog says its 50% launch offer lasts two weeks. The live API pricing page displayed the published Standard list rates on October 20, confirming the catalog's October 20 date-level transition; Mistral does not publish the exact cutoff time.
Under the expanded Microsoft–Mistral partnership announced 2026-07-21, Mistral Medium 3.5 is resold on Microsoft Foundry as a serverless listing, priced identically to Mistral's first-party API. Foundry publishes Global and Data Zone meters; no cached-input meter exists on either tier yet, so cache-heavy workloads get no discount on this model. (Mistral OCR 4 is billed per page rather than per token, so it is out of this catalog's scope.)
Pricing
per 1M tokens · USD / CHF| Model | Tier / Host | Context | Input | Cached | Output | Confidence |
|---|---|---|---|---|---|---|
Mistral Large 4 Public Preview Open (v26.10). Mistral's API pricing page displayed list rates on October 20, matching the catalog's date-level transition after the two-week launch offer. Batch and Priority use the same date boundary; the exact cutoff time is unpublished. | Direct API Mistral direct API | 1M | $1.36CHF 1.09 | $0.14CHF 0.113 | $4.18CHF 3.36 | official Mistral's official model card identifies the public-preview model ID `mistral-large-4` and 1M-token context. The live API pricing page, checked through its Standard, Batch, and Priority options on 2026-10-20, displays the published list rates. Mistral's October 6 changelog says the launch offer is 50% off for two weeks but does not publish an exact cutoff time; the catalog uses October 20 as the date-level boundary, 14 days after the release date. The full Azure Retail Prices feed checked 2026-10-20 has no Mistral Large 4 meter, so no Foundry per-token rate is inferred. |
List price (from October 20) List price (from October 20)$1.36 / $0.14 / $4.18Batch$0.34 / $0.035 / $1.04List price (from October 20)$0.68 / $0.07 / $2.09Priority$1.19 / $0.122 / $3.66List price (from October 20)$2.38 / $0.245 / $7.32 | ||||||
Mistral Medium 3.5 Foundry serverless Global tier; no cached-input meter published. | Foundry ·Global | — | $1.50CHF 1.21 | — | $7.50CHF 6.04 | official Azure Retail Prices API 'MM3.5 Inp/Outp glbl' meters at $0.0015/1K and $0.0075/1K ($1.50/M and $7.50/M), effective 2026-06-01 across 44 regions (including all 9 APAC regions); the only 2026-07-01 Global row is the malaysiawest region onboarding at the same price. Captured 2026-07-23, date corrected 2026-07-28. Announced via the Microsoft–Mistral partnership on 2026-07-21. No cached-input meter published on any tier. |
Mistral Medium 3.5 Data Zone tier at the usual ~10% premium over Global; no cached-input meter. | Foundry ·Data Zone | — | $1.65CHF 1.33 | — | $8.25CHF 6.64 | official Azure Retail Prices API 'MM3.5 Inp/Outp DZ' meters at $0.00165/1K and $0.00825/1K ($1.65/M and $8.25/M) — a clean 1.1x the Global rate — effective 2026-06-01, captured 2026-07-23. No cached-input meter published. |
Mistral Medium 3.5 mistral-medium-latest; identical to the Foundry Global rate. Cached input now officially published at $0.15/M (previously derived from Mistral's -90% cache discount rule). | Direct API Mistral direct API | — | $1.50CHF 1.21 | $0.15CHF 0.121 | $7.50CHF 6.04 | official Mistral's consolidated inference pricing page (docs.mistral.ai/inference/pricing), captured 2026-08-14: Mistral Medium 3.5 lists Input $1.5, Cached input $0.15, Output $7.5 per M (mode STANDARD/USD) — the cached rate is now a directly published dollar figure, superseding the earlier -90%-rule derivation from mistral.ai/pricing/api. Model card (docs.mistral.ai/models/mistral-medium-3-5-26-04) corroborates the $1.50/$7.50 input/output rate. Neither Foundry tier publishes a cached-input meter at all (re-confirmed against the Azure Retail Prices API in today's sweep), so those rows stay null. |
Quirks & gotchas
Foundry Global matches Mistral's direct API
Mistral Medium 3.5 costs the same on Microsoft Foundry Global ($1.50 / CHF 1.21 input, $7.50 / CHF 6.04 output) as on Mistral's own API — no Foundry resale markup at the Global tier. The Data Zone tier adds the usual ~10% ($1.65 / $8.25).
Foundry has no cache meter; the direct API's $0.15/M cache rate is now official
Neither Foundry tier publishes a cached-input meter for Mistral Medium 3.5, so Foundry workloads pay full input price on every call. The direct-API cached rate ($0.15/M) used to be derived as 10% of the $1.50 input price from a published -90% discount rule with no dollar figure; Mistral's new consolidated inference pricing page now publishes $0.15/M directly, so the number is unchanged but the confidence is upgraded to official. Models like Grok-4.3 ($0.20/M cached, official) or the Kimi K2 line ($0.10–$0.16/M cached, official) remain cheaper for cache-heavy use on Foundry.