Manufacturer

Google

Gemma-Modellfamilie von Google DeepMind.

Website ↗ 🤗 Hugging Face Filter in leaderboard

15Models

429Benchmarks

421Performance

8Harness

Best token generation

2.491,2 tok/s

gemma-4-E2B-it

Best prefill

55.259,6 tok/s

gemma-4-E2B-it

∅ Token generation

318,5 tok/s

Average of 421 runs

∅ Prefill

6.087,8 tok/s

Average of 421 runs

Best TTFT

91 ms

gemma-4-31B-it

Best harness rate

91%

Gemma-4-26B-A4B-it

🏆

gemma-4-E2B-it · fastest model at 2.491,2 tok/s token generation
Best prefill of this manufacturer: 55.259,6 tok/s (gemma-4-E2B-it)

Top models by token generation

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

gemma-4-E2B-it

2.491,2 tok/s

gemma-4-E4B-it

1.758,6 tok/s

Gemma-4-26B-A4B-it

1.589,1 tok/s

gemma-4-12B-it

1.104,3 tok/s

gemma-2-2b-it

323,8 tok/s

gemma-3n-E4B-it

273,1 tok/s

Gemma-4-26B-A4B

264,1 tok/s

gemma-4-31B-it

251,1 tok/s

Models

15 assigned

All benchmarks →

gemma-2-27b-it27B3 runs

69,4tok/s Ø

Min 31,9Max 113,0

gemma-2-2b-it2B9 runs

134,9tok/s Ø

Min 3,8Max 323,8

gemma-2-9b-it9B3 runs

113,3tok/s Ø

Min 56,5Max 171,6

gemma-3n-E4B-it8B9 runs

161,5tok/s Ø

Min 49,0Max 273,1

gemma-4-12B-it12B100 runs

253,7tok/s Ø

Min 22,3Max 1.104,3

gemma-4-12B-it-assistant12B0 runs

No performance data yet

Gemma-4-26B-A4B26B7 runs

155,4tok/s Ø

Min 47,1Max 264,1

Gemma-4-26B-A4B-it26B90 runs

372,4tok/s Ø

Min 32,4Max 1.589,1

gemma-4-26B-A4B-it-assistant26B3 runs

58,3tok/s Ø

Min 33,0Max 74,5

gemma-4-31B-it31B55 runs

78,6tok/s Ø

Min 10,5Max 251,1

gemma-4-31B-it-assistant31B0 runs

No performance data yet

gemma-4-E2B-it5B97 runs

482,2tok/s Ø

Min 6,8Max 2.491,2

gemma-4-E2B-it-assistant5B0 runs

No performance data yet

gemma-4-E4B-it8B53 runs

421,2tok/s Ø

Min 37,9Max 1.758,6

gemma-4-E4B-it-assistant8B0 runs

No performance data yet