The short answer
● Mistral Large 4 — 1T total, weights published today
● Inkling 2 Small — 594B total, weights still not released
● Mistral licence — modified MIT, MaaS needs a separate agreement
● Inkling licence — Apache 2.0 promised after safety testing, no date
● Available now — Mistral. Inkling runs only on Tinker
Two flagship releases in four days, both described as open weight. One of them you can download right now.
|
Mistral Large 4 |
Inkling 2 Small |
| Lab |
Mistral |
Thinking Machines |
| Announced |
9 November 2026 |
6 November 2026 |
| Parameters |
1T total, around 50B active |
594B total, 17B active |
| Weights available |
Yes, on Hugging Face |
No |
| Licence |
Modified MIT, MaaS excluded |
Apache 2.0 promised, undated |
| Where to run it |
Self-host, NIM, Foundry, OpenRouter |
Tinker platform only |
| Hosted price |
$0.70 / $2.10 per million |
$1.18 per task |
| Input |
Natively multimodal |
Text and image |
Two different meanings of open
Mistral published the weights and attached one commercial restriction. Thinking Machines announced an intention and attached a safety review with no end date. Both get described the same way in coverage, and only one of them is something you can act on this week.
That matters if you are planning rather than browsing. A promise of Apache 2.0 weights is worth having, but it is not a dependency you can schedule around.
If the restriction applies to you
Mistral’s exclusion is narrow and specific: businesses operating a Model-as-a-Service offering need a separate commercial licence. For most teams building a product on top of a model, that clause never fires. For anyone reselling inference, it fires immediately and Inkling’s eventual Apache 2.0 would be the cleaner licence — whenever it arrives.
On cost
The two are quoted in different units, which makes a direct comparison dishonest. Mistral publishes $0.70 input and $2.10 output per million tokens. Artificial Analysis puts Inkling 2 Small at $1.18 per task on roughly 150,000 output tokens. Your own token profile decides which is cheaper, and neither vendor has published the other’s metric.
Sources: Mistral AI · The Decoder.
Published 8 November 2026. Every figure above was checked against the reporting linked here on the same day.