Created bymario-alka.dePowered bygodcore.denoob2claw.detricoma.de
DEEN
Sign in Sign up
Beta
Open data · reproducible tests

Which model is fast & smart?

We benchmark AI language models on real hardware: speed in tokens per second and response time – plus quality on real agent tasks. Transparent, reproducible, directly comparable.

Start typing – or jump straight in:Fastest (tok/s)Best harnessAll modelsMakersHighscore
3863 benchmarks69 models2 test types
Live benchmark
Who generates text the fastest? Real measurements, real hardware.
1
gemma-4-E2B-itGeForce RTX 5090
2.491,2tok/s
2
NVIDIA-Nemotron-3-Nano-4BRTX PRO 6000 Blackwe
2.182,7tok/s
3
gpt-oss-20bRTX PRO 6000 Blackwe
1.995,4tok/s
4
Ornith-1.0-35BRTX PRO 6000 Blackwe
1.799,4tok/s
5
Nemotron-3-Nano-Omni-30B-A3B-ReasoningRTX PRO 6000 Blackwe
1.791,5tok/s
Peak2.491tok/s
View ranking →
01

Real performance

TTFT, prefill and generation tokens/s instead of theoretical peak numbers.

02

Harness quality

Points and success rates from complete agent and chat runs.

03

Full transparency

Model, quantization, runtime and hardware stay visible on every run.

Community favorites

Top models by benchmarks

Speed (tokens/s) and quality (harness) of the most-tested models – directly comparable.

All models →
🥇
Qwen (Alibaba)

Qwen3.6-35B-A3B

MoE35B
⚡ Speed1.489,1 tok/s
🏆 Quality–
137 Benchmarks137 Perf.0 Harness
🥈
Qwen (Alibaba)

Qwen3-30B-A3B-Instruct-2507

MoE30B
⚡ Speed1.671,0 tok/s
🏆 Quality–
131 Benchmarks131 Perf.0 Harness
🥉
Qwen (Alibaba)

Qwen3-30B-A3B-Thinking-2507

MoE30B
⚡ Speed1.683,8 tok/s
🏆 Quality–
126 Benchmarks126 Perf.0 Harness
4
Qwen (Alibaba)

Qwen3-Coder-30B-A3B-Instruct

MoE30B
⚡ Speed1.716,6 tok/s
🏆 Quality–
125 Benchmarks125 Perf.0 Harness
5
OpenAI

gpt-oss-20b

MoE20B
⚡ Speed1.995,4 tok/s
🏆 Quality–
124 Benchmarks124 Perf.0 Harness
6
Mistral AI

Magistral-Small-2509

Dense24B
⚡ Speed849,3 tok/s
🏆 Quality–
119 Benchmarks119 Perf.0 Harness
Current measurements

Leaderboard Snapshot

Alle Ergebnisse →
Metric:
Size
#Model / MakerMetricsParallelGPU / CPU / RAMRuntime
1gemma-4-E2B-it5BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation)
2.491,20 tok/s TG
Prefill 15.814 · TTFT 4.382 ms
10×NVIDIA GeForce RTX 5090AMD Ryzen 7 5800X3D 8-Core Processor · 126 GB RAMllama.cppcodex_cliQ4_K_MDetails →
2gemma-4-E2B-it5BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation)
2.414,37 tok/s TG
Prefill 20.277 · TTFT 4.107 ms
10×NVIDIA RTX PRO 6000 Blackwell Workstation EditionAMD Ryzen 9 9950X 16-Core Processor · 92 GB RAMllama.cppgodclawQ4_K_MDetails →
3NVIDIA-Nemotron-3-Nano-4B4BNVIDIA Performance benchmarkTimebench 3 - Kombi (Prefill + Generation)
2.182,67 tok/s TG
Prefill 10.599 · TTFT 5.598 ms
10×NVIDIA RTX PRO 6000 Blackwell Workstation EditionAMD Ryzen 9 9950X 16-Core Processor · 92 GB RAMllama.cppgodclawQ4_K_MDetails →
4gemma-4-E2B-it5BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation)
2.143,55 tok/s TG
Prefill 13.632 · TTFT 4.819 ms
10×3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation EditionAMD Ryzen Threadripper PRO 9965WX 24-Cores · 125 GB RAMllama.cppgodclawQ4_K_MDetails →
5gpt-oss-20b20BOpenAI Performance benchmarkTimebench 3 - Kombi (Prefill + Generation)
1.995,37 tok/s TG
Prefill 16.158 · TTFT 5.309 ms
10×NVIDIA RTX PRO 6000 Blackwell Workstation EditionAMD Ryzen 9 9950X 16-Core Processor · 92 GB RAMllama.cppgodclawQ4_K_MDetails →
Hardware ranking

Benchmarks by graphics card

Generation throughput (tokens/s) per GPU – minimum, average and maximum across all performance runs.

All runs →
NVIDIA GeForce RTX 5090415 Läufe
527,1tok/s Ø
Min 0,1Max 2.491,2
NVIDIA RTX PRO 6000 Blackwell Workstation Edition279 Läufe
473,8tok/s Ø
Min 0,0Max 2.414,4
NVIDIA GeForce RTX 3090 Ti466 Läufe
303,1tok/s Ø
Min 0,0Max 1.492,8
NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition1121 Läufe
292,4tok/s Ø
Min 0,1Max 2.143,6
NVIDIA GB1049 Läufe
169,9tok/s Ø
Min 11,4Max 903,3
NVIDIA GeForce RTX 5070 Ti237 Läufe
161,9tok/s Ø
Min 0,7Max 1.744,0
Intel Arc Pro B7032 Läufe
141,6tok/s Ø
Min 0,0Max 826,2
AMD Radeon 8060S Graphics292 Läufe
67,3tok/s Ø
Min 7,0Max 173,3
AMD Radeon AI PRO R9700795 Läufe
64,9tok/s Ø
Min 0,0Max 667,0
Hardware ranking

Benchmarks by CPU

Generation throughput (tokens/s) per CPU – minimum, average and maximum across all performance runs.

All runs →
AMD Ryzen 7 5800X3D 8-Core Processor415 Läufe
527,1tok/s Ø
Min 0,1Max 2.491,2
AMD Ryzen 9 9950X 16-Core Processor282 Läufe
469,2tok/s Ø
Min 0,0Max 2.414,4
AMD Ryzen 9 8945HX with Radeon Graphics418 Läufe
312,1tok/s Ø
Min 0,0Max 1.492,8
AMD Ryzen Threadripper PRO 9965WX 24-Cores1119 Läufe
292,9tok/s Ø
Min 0,1Max 2.143,6
AMD Ryzen 5 5600X 6-Core Processor48 Läufe
224,7tok/s Ø
Min 0,4Max 830,5
NVIDIA Grace48 Läufe
171,8tok/s Ø
Min 11,4Max 903,3
AMD Ryzen Threadripper PRO 5975WX 32-Cores237 Läufe
161,9tok/s Ø
Min 0,7Max 1.744,0
AMD Ryzen 9 7945HX with Radeon Graphics32 Läufe
141,6tok/s Ø
Min 0,0Max 826,2
AMD RYZEN AI MAX+ 395 w/ Radeon 8060S292 Läufe
67,3tok/s Ø
Min 7,0Max 173,3
AMD Ryzen Threadripper PRO 7955WX 16-Cores795 Läufe
64,9tok/s Ø
Min 0,0Max 667,0
Intel(R) Xeon(R) CPU E5-4620 v2 @ 2.60GHz38 Läufe
4,0tok/s Ø
Min 0,0Max 14,5
Intel(R) Xeon(R) CPU E5-4657L v2 @ 2.40GHz1 Läufe
2,3tok/s Ø
Min 2,3Max 2,3
Frisch gemessen

Neuste Benchmarks

Alle Ergebnisse →
Performance benchmark11.08.2026
Qwen3.5-122B-A10B
Qwen (Alibaba)
24,12 tok/s
Mario Alka
Performance benchmark11.08.2026
Qwen3.5-122B-A10B
Qwen (Alibaba)
17,52 tok/s
Mario Alka
Performance benchmark11.08.2026
Qwen3.5-122B-A10B
Qwen (Alibaba)
7,93 tok/s
Mario Alka
Performance benchmark11.08.2026
Laguna-M.1
Poolside
2,95 tok/s
Mario Alka
Performance benchmark10.08.2026
Laguna-M.1
Poolside
1,57 tok/s
Mario Alka
Performance benchmark10.08.2026
Laguna-M.1
Poolside
2,91 tok/s
Mario Alka
Performance benchmark10.08.2026
Qwen3-30B-A3B-Instruct-2507
Qwen (Alibaba)
143,56 tok/s
Mario Alka
Performance benchmark10.08.2026
Qwen3-30B-A3B-Instruct-2507
Qwen (Alibaba)
140,06 tok/s
Mario Alka
Performance benchmark10.08.2026
Qwen3-30B-A3B-Instruct-2507
Qwen (Alibaba)
78,10 tok/s
Mario Alka
Performance benchmark10.08.2026
Qwen3.5-122B-A10B
Qwen (Alibaba)
29,08 tok/s
Mario Alka
Performance benchmark10.08.2026
Qwen3.5-122B-A10B
Qwen (Alibaba)
23,94 tok/s
Mario Alka
Performance benchmark10.08.2026
Qwen3.5-122B-A10B
Qwen (Alibaba)
10,46 tok/s
Mario Alka
Performance benchmark10.08.2026
Tencent-Hy3-295B-A21B
Tencent
1,57 tok/s
Mario Alka
Performance benchmark10.08.2026
Laguna-M.1
Poolside
2,84 tok/s
Mario Alka
Performance benchmark10.08.2026
Llama-3_1-Nemotron-Ultra-253B-v1
NVIDIA
0,00 tok/s
Mario Alka
Performance benchmark10.08.2026
Tencent-Hy3-295B-A21B
Tencent
0,98 tok/s
Mario Alka
Performance benchmark10.08.2026
Laguna-M.1
Poolside
2,90 tok/s
Mario Alka
Performance benchmark10.08.2026
Llama-3_1-Nemotron-Ultra-253B-v1
NVIDIA
0,00 tok/s
Mario Alka
Performance benchmark09.08.2026
Llama-3_1-Nemotron-Ultra-253B-v1
NVIDIA
0,09 tok/s
Mario Alka
Performance benchmark09.08.2026
Llama-3_1-Nemotron-Ultra-253B-v1
NVIDIA
0,00 tok/s
Mario Alka
Performance benchmark09.08.2026
Tencent-Hy3-295B-A21B
Tencent
0,00 tok/s
Mario Alka
Performance benchmark09.08.2026
Llama-3_1-Nemotron-Ultra-253B-v1
NVIDIA
0,00 tok/s
Mario Alka
Performance benchmark09.08.2026
Laguna-M.1
Poolside
2,85 tok/s
Mario Alka
Performance benchmark09.08.2026
Tencent-Hy3-295B-A21B
Tencent
0,00 tok/s
Mario Alka
Performance benchmark09.08.2026
Laguna-S-2.1-INT4
Poolside
774,90 tok/s
Mario Alka