Devstral 2 2512
Dec 2025
Retired 2026-07-31Replace withMistral Medium 3.5Newer version
MiMo v2.6 ProCheaper, as good
GPT-6 LunaBetter
Qwen3.8 FlashOpen weights
Mistral Medium 3.5Newer versionMiMo v2.6 ProCheaper, as goodGPT-6 LunaBetterQwen3.8 FlashOpen weightsDevstral 2 2512Compare inon
Benchmarks of Devstral 2 251229
BenchmarkByScoreRank
Overall1
Knowledge4
Science3
Coding5
Agents4
Documents2
Legacy · saturated or no longer reported10
No benchmark matches
Rankingin non-thinking mode?
Retired models are left out of the current rankings.
Alternatives?
ModelScoreCostPer taskWhat changes
CheaperMiMo v2.6 ProthinkingOpen×5.4-32%$0.13Memory (4-bit): 75 GB → 615 GB+4BetterGPT-6 Lunalow×2.5-75%$0.005Function calling only with none · Chat Completions+3OpenQwen3.8 Flashthinking×4.6-71%$0.37Memory (4-bit): 75 GB → 108 GB+6
CheaperDeepSeek V4.1 Flash 1017maxOpen×5.2-34%$0.26Memory (4-bit): 75 GB → 458 GB+5
CheaperClaude Haiku 5.5max×5-75%$0.21Remove sampling parameters+5ModelScoreCostPer taskWhat changes
MiMo v2.6 ProthinkingOpen×5.4-32%$0.13Memory (4-bit): 75 GB → 615 GB+4DeepSeek V4.1 Flash 1017maxOpen×5.2-34%$0.26Memory (4-bit): 75 GB → 458 GB+5Claude Haiku 5.5max×5-75%$0.21Remove sampling parameters+5
MiniMax M3.1 Flashthinking×5-67%$0.22Complete model response must be appended
GLM 5.3 FlashmaxOpen×4.9-70%$0.28Memory (4-bit): 75 GB → 193 GB+5ModelScoreCostPer taskWhat changes
GPT-6 Lunalow×2.5-75%$0.005Function calling only with none · Chat Completions+3GPT-5.6 Lunalow×2.4-44%$0.01Mistral Small 4thinkingOpen+31%-67%$0.015MiMo v2.5thinkingOpen×2.5-78%$0.019Memory (4-bit): 75 GB → 186 GB
Granite 4.2 8BthinkingOpen+29%-87%$0.024Context: 256K → 128KModelScoreCostPer taskWhat changes
Qwen3.8 FlashthinkingOpen×4.6-71%$0.37Memory (4-bit): 75 GB → 108 GB+6MiMo v2.6 FlashthinkingOpen×4.4-78%$0.062Memory (4-bit): 75 GB → 186 GB+4Qwen3.8 27BxhighOpen×3.9+20%$1.01max_tokens range · QwenCloud+4
Nemotron 3.5 Super VLOpen×3.1-78%$0.19TInkling SmallthinkingOpen×3-20%$0.089Memory (4-bit): 75 GB → 160 GB+1ModelScoreCostPer taskWhat changes
Mistral Large 4thinking×4.5—$1.13Infinite whitespace stream+4Mistral Medium 3.5highOpen+65%—$0.50Memory (4-bit): 75 GB → 77 GBMistral Small 4thinkingOpen+31%—$0.015Mistral Large 3non-thinkingOpensame—$0.031Memory (4-bit): not verifiedMagistral Medium 1thinkingsame——Context: 256K → 40KDevstral Mediumnon-thinkingOpensame—$0.40 / $2Context: 256K → 128K+1Magistral Small 1thinkingOpensame—$0.70 / $1.16Context: 256K → 128KMistral Small 3non-thinkingOpen-23%—$0.05 / $0.08Context: 256K → 32KAll Mistral models →Run it yourself
estimateFits on✗ 8 GB (phone, laptop)✗ 16 GB✗ 24 GB GPU✗ 48 GB GPU✓ 80 GB GPU · compressed
Licencecustom licencecommercial use with conditions
Benchmarks
all categoriesChat & Knowledge
AA Omniscience Index-46.7
GPQA Diamond59.4
Coding
SciCode32.8
Agents & Tool Use
τ³-Banking · AA10.5
τ²-bench · AA24.9
Writing & Language
FAQ
How much does Devstral 2 2512 cost?›
$0.40 per 1M input tokens and $2 per 1M output tokens; cached input $0.04 per 1M.
Is Devstral 2 2512 free to use?›
There is no free API; the cheapest paid price is $0.40 per 1M input tokens. The weights are open, so you can run it yourself.
How can I use Devstral 2 2512?›
Through the API of Mistral.
What is the context window of Devstral 2 2512?›
256K tokens, with up to 210K output tokens.
When was Devstral 2 2512 released? Is there a newer version?›
Released 2025-12-09. It has been superseded by Mistral Medium 3.5.
Is Devstral 2 2512 open source?›
Yes, the weights are open.
Can I run Devstral 2 2512 locally?›
The weights are open and can be self-hosted; hardware needs are not confirmed yet.
Can Devstral 2 2512 read images?›
It accepts text and PDF files as input.
What is Devstral 2 2512?›
Devstral 2 2512 is an open-weights language model from Mistral, released Dec 2025.
What are the best alternatives to Devstral 2 2512?›
MiMo v2.6 Pro (cheaper), GPT-6 Luna (better), Qwen3.8 Flash (open weights).