Value ranking
Best Value LLMs for Math in October 2026
As of October 2026, gpt-oss-20b gives the most math performance per dollar: 39.4 points at a blended $0.04 per million tokens. gpt-oss-120b is next.
Last verified
Category score per dollar of blended API price, among models in the top half of the math ranking.
- gpt-oss-20b 1095.0 pts/$
- gpt-oss-120b 747.2 pts/$
- Qwen3.7 Flash 696.0 pts/$
- Gemma 4 26B A4B IT 445.0 pts/$
- Nemotron 3 Nano 30B A3B 431.4 pts/$
- Nemotron 3.5 Lightning 428.3 pts/$
- GPT-6 Luna 380.5 pts/$
- Claude Haiku 5.5 367.9 pts/$
- MiMo-V2.6-Flash 296.7 pts/$
- Step 3.5 Flash 284.2 pts/$
Full ranking
| # | Model | Score | Input $/M | Output $/M | Points per $ |
|---|---|---|---|---|---|
| 1 | gpt-oss-20b | 39.4 | $0.018 | $0.09 | 1095.0 |
| 2 | gpt-oss-120b | 52.5 | $0.037 | $0.17 | 747.2 |
| 3 | Qwen3.7 Flash | 38.3 | $0.03 | $0.13 | 696.0 |
| 4 | Gemma 4 26B A4B IT | 47.6 | $0.0675 | $0.23 | 445.0 |
| 5 | Nemotron 3 Nano 30B A3B | 37.8 | $0.05 | $0.20 | 431.4 |
| 6 | Nemotron 3.5 Lightning | 37.5 | $0.05 | $0.20 | 428.3 |
| 7 | GPT-6 Luna | 76.1 | $0.10 | $0.50 | 380.5 |
| 8 | Claude Haiku 5.5 | 73.6 | $0.10 | $0.50 | 367.9 |
| 9 | MiMo-V2.6-Flash | 51.9 | $0.14 | $0.28 | 296.7 |
| 10 | Step 3.5 Flash | 42.6 | $0.10 | $0.30 | 284.2 |
| 11 | Gemma 4 31B IT | 43.2 | $0.09 | $0.34 | 283.1 |
| 12 | Hy3 | 40.1 | $0.0825 | $0.33 | 278.0 |
| 13 | DeepSeek V4.1 Flash | 66.7 | $0.15 | $0.60 | 254.2 |
| 14 | DeepSeek V4 Flash | 60.3 | $0.15 | $0.60 | 229.6 |
| 15 | Nemotron 3 Super | 39.6 | $0.08 | $0.45 | 229.4 |
| 16 | GLM-5.3-Flash | 53.3 | $0.15 | $0.50 | 224.2 |
| 17 | MiMo-V2-Omni | 39.1 | $0.14 | $0.28 | 223.4 |
| 18 | MiMo-V2-Flash | 38.3 | $0.14 | $0.28 | 219.0 |
| 19 | Gemini 2.5 Flash-Lite | 38.0 | $0.10 | $0.40 | 216.9 |
| 20 | Qwen3.5-Flash | 37.4 | $0.10 | $0.40 | 213.9 |
| 21 | MiMo-V2.5 | 36.8 | $0.14 | $0.28 | 210.0 |
| 22 | Qwen3-30B-A3B | 37.4 | $0.12 | $0.50 | 174.1 |
| 23 | GPT-5.6 Luna | 77.7 | $0.20 | $1.20 | 172.6 |
| 24 | DeepSeek-V3.2-Exp | 41.7 | $0.26 | $0.38 | 143.8 |
| 25 | Mistral Large 3 | 38.7 | $0.25 | $0.75 | 103.3 |
| 26 | Step 3.7 Flash | 42.9 | $0.18 | $1.11 | 103.1 |
| 27 | MiMo-V2.6-Pro | 54.5 | $0.43 | $0.87 | 100.2 |
| 28 | Trinity Large Thinking | 37.6 | $0.25 | $0.80 | 97.2 |
| 29 | Nvidia Llama 3.3 Nemotron Super 49b v1.5 | 38.2 | $0.40 | $0.40 | 95.6 |
| 30 | Qwen3.6 Flash | 39.0 | $0.19 | $1.13 | 92.5 |
| 31 | DeepSeek-V3.1 | 38.9 | $0.25 | $0.95 | 91.4 |
| 32 | GPT-5.4 nano | 40.9 | $0.20 | $1.25 | 88.4 |
| 33 | DeepSeek-V3.1-Terminus | 38.5 | $0.27 | $1 | 85.0 |
| 34 | MiniMax-M3 | 40.0 | $0.30 | $1.20 | 76.3 |
| 35 | Solar Pro4 | 38.8 | $0.30 | $1.20 | 73.8 |
| 36 | MiMo-V2.5-Pro | 40.0 | $0.43 | $0.87 | 73.6 |
| 37 | MiniMax-M2.1 | 38.3 | $0.30 | $1.20 | 73.0 |
| 38 | MiMo-V2-Pro | 39.5 | $0.43 | $0.87 | 72.6 |
| 39 | Gemini 3.1 Flash Lite | 40.7 | $0.25 | $1.50 | 72.3 |
| 40 | Qwen3.7 Plus | 50.5 | $0.40 | $1.60 | 72.1 |
| 41 | MiniMax-M2 | 37.3 | $0.30 | $1.20 | 71.1 |
| 42 | Inkling-Small | 45.1 | $0.45 | $1.20 | 70.7 |
| 43 | Qwen3.6 35B-A3B | 38.9 | $0.25 | $1.49 | 69.8 |
| 44 | GPT-5 Mini | 46.7 | $0.25 | $2 | 67.9 |
| 45 | DeepSeek V4 Pro | 64.8 | $0.66 | $1.98 | 65.5 |
| 46 | Qwen3 14B | 38.6 | $0.35 | $1.40 | 63.1 |
| 47 | Qwen3.5 35B-A3B | 39.9 | $0.25 | $2 | 58.0 |
| 48 | Kimi K2.5 | 51.8 | $0.45 | $2.25 | 57.6 |
| 49 | Qwen3.5 Plus | 49.6 | $0.40 | $2.40 | 55.1 |
| 50 | Hy4 preview | 55.7 | $0.75 | $2.25 | 49.5 |
| 51 | DeepSeek-R1 | 43.8 | $0.50 | $2.15 | 48.0 |
| 52 | Qwen3.5 27B | 38.8 | $0.30 | $2.40 | 47.0 |
| 53 | Gemini 2.5 Flash | 39.9 | $0.30 | $2.50 | 46.9 |
| 54 | Gemini 3.7 Flash | 69.6 | $0.75 | $3.75 | 46.4 |
| 55 | Qwen3.6 Plus | 51.8 | $0.50 | $3 | 46.0 |
| 56 | Gemini 3 Flash Preview | 51.7 | $0.50 | $3 | 46.0 |
| 57 | Qwen3-Next 80B-A3B Instruct | 38.8 | $0.50 | $2 | 44.3 |
| 58 | Nova 2 Lite | 37.5 | $0.30 | $2.50 | 44.1 |
| 59 | Gemini 3.8 Flash | 65.3 | $0.75 | $3.75 | 43.6 |
| 60 | Kimi K2 (Jul 2025) | 42.7 | $0.57 | $2.30 | 42.6 |
| 61 | GLM-4.5V | 37.4 | $0.60 | $1.80 | 41.5 |
| 62 | Qwen3 235B-A22B | 50.4 | $0.70 | $2.80 | 41.1 |
| 63 | Mistral Large 4 | 40.4 | $0.68 | $2.09 | 39.2 |
| 64 | GLM-4.6 | 39.1 | $0.60 | $2.20 | 39.1 |
| 65 | GLM-4.5 | 39.0 | $0.60 | $2.20 | 39.0 |
| 66 | MiniMax M1 | 37.5 | $0.55 | $2.20 | 39.0 |
| 67 | GLM-4.7 | 38.6 | $0.60 | $2.20 | 38.6 |
| 68 | Gemini 3.6 Flash | 57.3 | $0.75 | $3.75 | 38.2 |
| 69 | Muse Spark 1.3 | 73.1 | $1.25 | $4.25 | 36.6 |
| 70 | Qwen3.6 27B | 48.5 | $0.60 | $3.60 | 35.9 |
| 71 | Qwen3.5 122B-A10B | 39.1 | $0.40 | $3.20 | 35.6 |
| 72 | Seed 2.0 Pro | 39.3 | $0.50 | $3 | 34.9 |
| 73 | Qwen3.5 397B-A17B | 46.1 | $0.60 | $3.60 | 34.1 |
| 74 | Kimi K2.6 | 57.0 | $0.95 | $4 | 33.3 |
| 75 | Qwen3.8 27B | 37.1 | $0.99 | $1.49 | 33.2 |
| 76 | Qwen3 32B | 39.7 | $0.70 | $2.80 | 32.4 |
| 77 | Step 5 Preview | 46.1 | $1 | $2.70 | 32.4 |
| 78 | Qwen3-VL 235B-A22B | 39.0 | $0.70 | $2.80 | 31.8 |
| 79 | Kimi K2.7 Code | 52.9 | $0.95 | $4 | 30.9 |
| 80 | Grok 4.20 (Non-Reasoning) | 48.2 | $1.25 | $2.50 | 30.8 |
| 81 | GLM-5 | 46.4 | $1 | $3.20 | 29.9 |
| 82 | Grok 4.3 | 46.0 | $1.25 | $2.50 | 29.4 |
| 83 | GLM-5.3 | 62.3 | $1.40 | $4.40 | 29.0 |
| 84 | GPT-5.4 mini | 45.5 | $0.75 | $4.50 | 27.0 |
| 85 | GLM-5.2 | 55.7 | $1.40 | $4.40 | 25.9 |
| 86 | Grok 4.20 Multi-Agent | 39.4 | $1.25 | $2.50 | 25.2 |
| 87 | Qwen3.8 Max | 73.2 | $2 | $6 | 24.4 |
| 88 | GPT-6.1 Sol | 93.7 | $2 | $10 | 23.4 |
| 89 | Muse Spark 1.2 | 46.4 | $1.25 | $4.25 | 23.2 |
| 90 | GLM-5.1 | 49.7 | $1.40 | $4.40 | 23.1 |
| 91 | Muse Spark 1.1 | 45.5 | $1.25 | $4.25 | 22.8 |
| 92 | Claude Haiku 4.5 | 44.9 | $1 | $5 | 22.4 |
| 93 | Grok 4.6 | 67.0 | $2 | $6 | 22.3 |
| 94 | Claude Sonnet 5.5 | 87.9 | $2 | $10 | 22.0 |
| 95 | GPT-6 Sol | 87.2 | $2 | $10 | 21.8 |
| 96 | o4-mini | 40.8 | $1.10 | $4.40 | 21.2 |
| 97 | GLM-5V-Turbo | 39.4 | $1.20 | $4 | 20.7 |
| 98 | Grok 4.5 | 60.9 | $2 | $6 | 20.3 |
| 99 | Grok 4.7 | 57.8 | $2 | $6 | 19.3 |
| 100 | Qwen3.6 Max Preview | 54.1 | $1.30 | $7.80 | 18.5 |
| 101 | GPT-5.6 Terra | 81.6 | $2 | $12 | 18.1 |
| 102 | Gemini 3.5 Flash | 60.7 | $1.50 | $9 | 18.0 |
| 103 | Qwen3.7 Max | 62.4 | $2.50 | $7.50 | 16.7 |
| 104 | Claude Sonnet 5 | 66.2 | $2 | $10 | 16.6 |
| 105 | Qwen3 Max | 38.7 | $1.20 | $6 | 16.1 |
| 106 | GPT-5 | 55.0 | $1.25 | $10 | 16.0 |
| 107 | GPT-5.1 | 52.2 | $1.25 | $10 | 15.2 |
| 108 | o3 | 50.2 | $2 | $8 | 14.3 |
| 109 | Gemini 3.1 Pro Preview | 62.1 | $2 | $12 | 13.8 |
| 110 | GPT-5.4 | 73.5 | $2.50 | $15 | 13.1 |
| 111 | Mistral Medium 3.5 | 39.1 | $1.50 | $7.50 | 13.0 |
| 112 | Qwen3-Coder 480B-A35B Instruct | 37.6 | $1.50 | $7.50 | 12.5 |
| 113 | GPT-5.2 | 60.0 | $1.75 | $14 | 12.5 |
| 114 | Kimi K3 | 74.2 | $3 | $15 | 12.4 |
| 115 | Claude Opus 5.5 | 91.8 | $4 | $20 | 11.5 |
| 116 | GPT-5.6 Sol | 85.6 | $4 | $20 | 10.7 |
| 117 | Claude Sonnet 4.6 | 52.9 | $3 | $15 | 8.8 |
| 118 | Claude Opus 5 | 86.2 | $5 | $25 | 8.6 |
| 119 | GPT-5.3 Chat | 38.2 | $1.75 | $14 | 7.9 |
| 120 | Claude Opus 4.8 | 78.4 | $5 | $25 | 7.8 |
| 121 | GPT-5.5 | 81.7 | $5 | $30 | 7.3 |
| 122 | Claude Sonnet 4 | 43.3 | $3 | $15 | 7.2 |
| 123 | Claude Opus 4.7 | 66.7 | $5 | $25 | 6.7 |
| 124 | Claude Opus 4.6 | 63.0 | $5 | $25 | 6.3 |
| 125 | GPT-6 Astra | 93.5 | $10 | $50 | 4.7 |
| 126 | Claude Fable 5.1 | 89.6 | $10 | $50 | 4.5 |
| 127 | Claude Fable 5 | 88.5 | $10 | $50 | 4.4 |
| 128 | Claude Opus 4.5 | 38.6 | $5 | $25 | 3.9 |
| 129 | Claude Opus 4 | 42.0 | $15 | $75 | 1.4 |
| 130 | GPT-5.5 Pro | 84.0 | $30 | $180 | 1.2 |
| 131 | GPT-5 Pro | 48.5 | $15 | $120 | 1.2 |
| 132 | GPT-5.2 Pro | 65.3 | $21 | $168 | 1.1 |
| 133 | GPT-5.4 Pro | 72.4 | $30 | $180 | 1.1 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Compare the leaders
Frequently asked questions
What is the most cost-effective AI model for math?
As of October 2026, gpt-oss-20b gives the most math performance per dollar: 39.4 points at a blended $0.04 per million tokens. gpt-oss-120b is next.
What are the top 5 in this ranking?
gpt-oss-20b, gpt-oss-120b, Qwen3.7 Flash, Gemma 4 26B A4B IT, Nemotron 3 Nano 30B A3B, in that order, as of October 2026.
How is this list ranked?
By score per dollar: the model’s score divided by its blended API price (three parts input to one part output), among models in the top half by score.