Last verified
RETIRED SEP 30 2026LEAN 4 PROOFS256K CONTEXT119B / 6.5B ACTIVEAPACHE 2.0MISTRAL LABS

Leanstral 1.5 API Pricing

Retired September 30, 2026: Mistral deprecated labs-leanstral-1-5 on September 29 and retired it the next day, naming no replacement. Leanstral 1.5 was Mistral's model for Lean 4 formal proofs - automated theorem proving and autoformalization - and it cost nothing to call: a public preview under the Labs label, listed free from June 30, 2026, the day the earlier Leanstral 2603 was retired. The weights are Apache 2.0, so self-hosting remains an option.

Input - per 1M tokens
Free/M
Source mistral.ai free
Output - per 1M tokens
Free/M
Max output 128K free
Cached input
Not published
Cache nothing to discount not listed
Effective - agentic blend
Free/M
92/8 split - free row
§ 01 / TERMINAL

Run the numbers.

Calculator pre-loaded with Leanstral 1.5's listed rate, which is zero. It is here for the comparison: switch the model to see what the same proof workload would cost on a paid Mistral or third-party model.

$ /mo
Workload split
Tokens you can process
—
Words equivalent (English)
—
Effective rate
—
Open full calculator (all models · share URL · CSV) →
§ 02 / SCENARIOS

Real-world presets.

§ 03 / TOKENIZER

Paste text. See tokens. See cost.

What's a token →
Calibrated · measured on the vendor's tokenizer · 2026-06-10 Auto-counts as you type

Counts use a chars-per-token calibration measured on the vendor's own published tokenizer (mistralai/Mistral-Nemo-Instruct-2407, 2026-06-10). English prose is typically within a few percent; code and non-Latin scripts tokenize heavier. For billing-exact counts use the vendor's count-tokens API.

Characters 548
Words 95
Tokens (estimated) 108 tokens
Cost as input · uncached $0 USD
Cost as output · uncached $0 USD
§ 04 / SHELF

Up against the shelf.

All models →
Model Input /M Output /M Effective blended Context Best for
Leanstral 1.5 Current Free Free Free agentic 92/8 256K Lean 4 formal proofs, free until retired
Leanstral Free Free Free retired documented elsewhere The Leanstral this model replaced
Codestral $0.30 cache $0.03 $0.90 $0.144 paid documented elsewhere — not on pricing page Mistral's paid coding model
Mistral Medium 3.5 $1.50 cache $0.15 $7.50 $0.962 paid documented elsewhere — not on pricing page Frontier agentic and coding work
Mistral Small 4 $0.15 cache $0.015 $0.60 $0.0842 paid 256K Low-cost reasoning and coding
DeepSeek V4.1 Flash $0.30 cache $0.006 $1.20 $0.15 paid 1M Budget coding and bulk reasoning

Frequently asked.

Leanstral 1.5 pricing questions - what free means here, and what it does not promise.

Q · 01 Can I still call Leanstral 1.5? +
No. Mistral deprecated labs-leanstral-1-5 on September 29, 2026 and retired it on September 30, naming no replacement - its deprecated and retired models table leaves the alternative column empty. Mistral's API pricing page still listed the model as free on the retirement day, but the lifecycle table is what governs the endpoint. The weights are Apache 2.0, so self-hosting is the remaining route.
Q · 02 Was it really free? +
Yes, while it ran. The model page showed PRICE: FREE and the API pricing page listed the Leanstral endpoint as Free, kept open, in Mistral's words, "for a limited period to gather realistic feedback". It was a Labs preview, never a published long-term price.
Q · 03 What model IDs did Leanstral use? +
labs-leanstral-1-5 from June 30 to September 30, 2026, on chat completions, agents and conversations. The earlier labs-leanstral-2603 was retired on June 30, 2026. Both ids are now retired.
Q · 04 What was it for? +
Lean 4 formal proof engineering: automated theorem proving and autoformalization, turning informal statements into Lean. Mistral lists function calling and structured outputs, so it can drive a proof loop that calls Lean's checker as a tool. It is not a general chat or coding model - for that Mistral sells Codestral and Medium 3.5.
Q · 05 How big is the context window? +
256K tokens, with up to 128K output tokens per request. Proof search tends to produce long outputs, so the output ceiling matters more here than on most chat models.
Q · 06 Can I self-host it? +
Mistral lists Leanstral 1.5 under the Apache 2.0 license, and with the API endpoint retired, self-hosting is the way to keep using it. A 119B-parameter mixture-of-experts model (6.5B active) costs whatever the hardware costs; the free price on this page applied to Mistral's API only.
Q · 07 What happened to the original Leanstral? +
Leanstral 2603, also free, was retired on June 30, 2026. Leanstral 1.5 is dated the same day on Mistral's model page, so there was no gap in free Lean 4 access - only a model-id change.