Cheapest good LLM for math
- DeepSeek V4 Flash 0731 leads
- $0.396
- models qualify
- 54
Output $/Mtok · read October 22
Cheapest output, $ per million tokens
- S Llama 3 8B Lunaris $0.05
- Gemma 3 4B $0.075
- Mistral Small 3 $0.08
- Llama 3.1 8B Instruct $0.08
- N Nex-N2.5-Mini $0.10
- Reka Edge $0.10
- Ministral 3 3B 2512 $0.10
- Granite 4.0 Micro $0.112
As of October 22, 2026, DeepSeek V4 Flash 0731 is the cheapest model at or above the median Math Index result, at $0.396 per million output tokens, ahead of Google Gemma 4 31B ($0.40) and Anthropic Claude Haiku 5.5 ($0.50).
The full best LLM for math ranking combines this benchmark with others.
| Rank | Model | Output $/Mtok | Coding evidence | Math Index | Coding index |
|---|---|---|---|---|---|
| 1 | DeepSeek V4 Flash 0731 DeepSeek | $0.396 | 76.0 | 51.8 | |
| 2 | Gemma 4 31B Google | $0.40 | 74.6 | 39.0 | |
| 3 | Claude Haiku 5.5 Anthropic | $0.50 | 87.0 | 70.7 | |
| 4 | GPT-6 Luna OpenAI | $0.50 | 84.8 | 57.3 | |
| 5 | gpt-oss-120b OpenAI | $0.50 | 80.9 | Not enough data | |
| 6 | GLM 5.3 Flash Z.ai | $0.50 | 76.6 | 57.7 | |
| 7 | DeepSeek V4.1 Flash 1017 DeepSeek | $0.60 | 85.3 | 70.3 | |
| 8 | Gemma 4 26B A4B Google | $0.60 | 80.9 | Not enough data | |
| 9 | DeepSeek V4.1 Flash DeepSeek | $0.84 | 82.2 | 62.5 | |
| 10 | GPT-5.6 Luna OpenAI | $1.20 | 87.5 | 58.0 | |
| 11 | DeepSeek V4 Pro 0813 DeepSeek | $1.98 | 80.0 | 56.1 | |
| 12 | GLM 4.7 Z.ai | $2.20 | 79.1 | 55.1 | |
| 13 | Nemotron 3 Ultra NVIDIA | $2.40 | 83.2 | 30.0 | |
| 14 | Gemini 3 Flash Preview Google | $3.00 | 73.4 | 43.7 | |
| 15 | GLM 5 Z.ai | $3.20 | 77.8 | 41.7 | |
| 16 | $3.50 | 80.7 | Not benchmarked | ||
| 17 | Gemini 3.8 Flash Google | $3.75 | 84.7 | 61.6 | |
| 18 | Gemini 3.7 Flash Google | $3.75 | 84.5 | 58.0 | |
| 19 | Gemini 3.6 Flash Google | $3.75 | 78.2 | 62.5 | |
| 20 | Kimi K2.6 Moonshot AI | $4.00 | 77.1 | 52.1 |
How this list is built
Generated from the dataset on every refresh. Nobody picks the order by hand.
Models whose Math Index result is at or above 72.3, the median of every model on that board, ordered by output price, cheapest first. Free listings are left out: they are a tariff, not a price.
- OpenRouter, fetched
- Epoch AI benchmarking hub, fetched
- LMArena (Hugging Face dataset), fetched