Models

Params B is total weight count; Active B is what actually runs per token on a mixture-of-experts model (equal to Params B otherwise). Measured counts the independent benchmark records run for that model, across all hardware.

ModelFamilyArchitectureParams BActive BReleasedMeasuredBest tok/sPriceEvidence
DeepSeek R1 (671B-A37B)deepseekMoE671372025-01-20710.7$9850 (new)single-source: One credible source
DeepSeek R1 Distill Llama 8BdeepseekDense8.038.032025-01-20150$11700 (used)single-source: One credible source
DeepSeek R1 Distill Qwen 32BdeepseekDense32.7632.762025-01-20117$3499 (new)single-source: One credible source
DeepSeek R1-0528 (671B-A37B)deepseekMoE671372025-05-2816$11200 (used)single-source: One credible source
DeepSeek V3 (671B-A37B)deepseekMoE671372024-12-2516.2$20600 (used)single-source: One credible source
Gemma 2 9B Instructgemma2Dense9.249.242024-06-240
Gemma 3 12B Instructgemma3Dense12.1911.772025-03-120
Gemma 3 27B Instructgemma3Dense27.4327.012025-03-12135.2$20600 (used)single-source: One credible source
Gemma 3 4B Instructgemma3Dense4.33.882025-03-12381.1$4699 (new)single-source: One credible source
Gemma 4 E2B Instructgemma4Dense5.122.32026-03-0216$145 (new)single-source: One credible source
gpt-oss-120bgpt-ossMoE116.835.12025-08-051060.6$4699 (new)single-source: One credible source
gpt-oss-20bgpt-ossMoE21.513.62025-08-0510147.5$1274 (used)single-source: One credible source
Llama 2 7Bllama2Dense6.746.742023-07-1381287.6$3400 (used)verified: Two or more independent sources agreeing within 25%
Llama 3 8B Instructllama3Dense8.038.032024-04-181122.6$5000 (used)single-source: One credible source
Llama 3.1 8B Instructllama3Dense8.038.032024-07-233144.7$1274 (used)single-source: One credible source
Llama 3.2 1B Instructllama3Dense1.241.242024-09-250
Llama 3.2 3B Instructllama3Dense3.213.212024-09-255231$20600 (used)single-source: One credible source
Llama 3.3 70B Instructllama3Dense70.5570.552024-12-0627.1$11700 (used)single-source: One credible source
Llama 4 Scout 17B-16E Instructllama4MoE108.64172025-04-050
LLaMA 7BllamaDense6.746.742023-02-24153.5$2750 (new)single-source: One credible source
Mistral 7B Instruct v0.2mistralDense7.247.242023-12-11139.7$1120 (new)single-source: One credible source
Mistral 7B Instruct v0.3mistralDense7.257.252024-05-221130.7$20600 (used)single-source: One credible source
Mistral Small 3 24B Instruct (2501)mistralDense23.5723.572025-01-300
Mistral Small 3.2 24B Instruct (2506)mistralDense24.0123.572025-06-190
Mixtral 8x22B v0.1mistralMoE140.6239.152024-04-1613.9$11650 (used)single-source: One credible source
Mixtral 8x7B Instruct v0.1mistralMoE46.712.92023-12-100
Phi-4 14BphiDense14.6614.662024-12-110
Qwen2.5 14B Instructqwen2.5Dense14.7714.772024-09-160
Qwen2.5 32B Instructqwen2.5Dense32.7632.762024-09-170
Qwen2.5 72B Instructqwen2.5Dense72.7172.712024-09-16111.5single-source: One credible source
Qwen2.5 7B Instructqwen2.5Dense7.627.622024-09-16122.6$799 (new)single-source: One credible source
Qwen3 0.6Bqwen3Dense0.750.752025-04-29154.3$399 (new)single-source: One credible source
Qwen3 1.7Bqwen3Dense2.032.032025-04-281161.4$4699 (new)single-source: One credible source
Qwen3 14Bqwen3Dense14.7714.772025-04-29252.1$1274 (used)single-source: One credible source
Qwen3 235B-A22Bqwen3MoE235.09222025-04-290
Qwen3 235B-A22B Instruct 2507qwen3MoE235.09222025-07-210
Qwen3 235B-A22B Thinking 2507qwen3MoE235.09222025-07-250
Qwen3 30B-A3Bqwen3MoE30.533.32025-04-294153.6$1274 (used)single-source: One credible source
Qwen3 30B-A3B Instruct 2507qwen3MoE30.533.32025-07-287113.3$3499 (new)single-source: One credible source
Qwen3 30B-A3B Thinking 2507qwen3MoE30.533.32025-07-300
Qwen3 32Bqwen3Dense32.7632.762025-04-295879.3single-source: One credible source
Qwen3 4Bqwen3Dense4.024.022025-04-290
Qwen3 4B Instruct 2507qwen3Dense4.024.022025-08-068134.5$3499 (new)single-source: One credible source
Qwen3 4B Thinking 2507qwen3Dense4.024.022025-08-060
Qwen3 8Bqwen3Dense8.198.192025-04-2912115.3$1274 (used)single-source: One credible source
Qwen3 Coder 30B-A3B Instructqwen3MoE30.533.32025-07-31373.1$199 (used)single-source: One credible source
Qwen3 Coder 480B-A35B Instructqwen3MoE480.15352025-07-220
Qwen3 Coder Next (80B-A3B)qwen3MoE79.6732026-01-300
Qwen3 Next 80B-A3B Instructqwen3MoE81.3232025-09-090
Qwen3 Next 80B-A3B Thinkingqwen3MoE81.3232025-09-090
Qwen3.5 0.8Bqwen3.5Dense0.870.872026-03-010
Qwen3.5 122B-A10Bqwen3.5MoE125.09102026-02-24138.7$4699 (new)verified: Two or more independent sources agreeing within 25%
Qwen3.5 27Bqwen3.5Dense27.7827.782026-02-240
Qwen3.5 2Bqwen3.5Dense2.272.272026-03-010
Qwen3.5 35B-A3Bqwen3.5MoE35.9532026-02-24350.3$4699 (new)single-source: One credible source
Qwen3.5 397B-A17Bqwen3.5MoE403.4172026-02-160
Qwen3.5 4Bqwen3.5Dense4.664.662026-02-270
Qwen3.5 9Bqwen3.5Dense9.659.652026-02-270
Qwen3.6 27Bqwen3.6Dense27.7827.782026-04-21414.7single-source: One credible source
Qwen3.6 35B-A3Bqwen3.6MoE35.9532026-04-15878.8single-source: One credible source
Qwen3.8 27Bqwen3.8Dense27.7827.782026-08-14447.4$2500 (used)verified: Two or more independent sources agreeing within 25%