Manufacturer

Mistral AI

Franzoesisches KI-Labor; Mistral, Mixtral, Codestral, Devstral, Ministral, Magistral.

Website ↗ 🤗 Hugging Face Filter in leaderboard

10Models

640Benchmarks

636Performance

4Harness

Best token generation

1.060,0 tok/s

Ministral-3-14B-Reasoning-2512

Best prefill

20.061,1 tok/s

Ministral-3-14B-Reasoning-2512

∅ Token generation

140,5 tok/s

Average of 636 runs

∅ Prefill

3.123,1 tok/s

Average of 636 runs

Best TTFT

20 ms

Magistral-Small-2509

Best harness rate

87%

Devstral-Small-2-24B-Instruct-2512

🏆

Ministral-3-14B-Reasoning-2512 · fastest model at 1.060,0 tok/s token generation
Best prefill of this manufacturer: 20.061,1 tok/s (Ministral-3-14B-Reasoning-2512)

Top models by token generation

Best measured generation throughput per model (tok/s), published performance benchmarks.

Ministral-3-14B-Reasoning-2512

1.060,0 tok/s

Mistral-Small-4-119B-2603

959,6 tok/s

Magistral-Small-2509

737,0 tok/s

Devstral-Small-2-24B-Instruct-2512

725,6 tok/s

Devstral-Small-2507

647,6 tok/s

Codestral-22B-v0.1

641,4 tok/s

Mistral-Small-3.1-24B-Instruct-2503

214,9 tok/s

Devstral-2-123B-Instruct-2512

143,1 tok/s

Models

10 assigned

All benchmarks →

Codestral-22B-v0.122B89 runs

129,0tok/s Ø

Min 13,0Max 641,4

Devstral-2-123B-Instruct-2512123B50 runs

22,1tok/s Ø

Min 0,0Max 143,1

Devstral-Small-2-24B-Instruct-251224B94 runs

156,1tok/s Ø

Min 8,9Max 725,6

Devstral-Small-250724B86 runs

150,3tok/s Ø

Min 13,3Max 647,6

Magistral-Small-250924B92 runs

172,9tok/s Ø

Min 4,0Max 737,0

Mamba-Codestral-7B-v0.17B0 runs

No performance data yet

Ministral-3-14B-Reasoning-251214B106 runs

253,1tok/s Ø

Min 22,4Max 1.060,0

Mistral-Medium-3.5-128B128B28 runs

10,2tok/s Ø

Min 0,6Max 141,6

Mistral-Small-3.1-24B-Instruct-250324B69 runs

63,2tok/s Ø

Min 5,0Max 214,9

Mistral-Small-4-119B-2603119B26 runs

93,1tok/s Ø

Min 5,8Max 959,6