Cheapest good LLM for reasoning
- Ling 3.0 Flash leads
- $0.063
- models qualify
- 46
Output $/Mtok · read October 21
Cheapest output, $ per million tokens
- Mistral Nemo $0.03
- Gemma 3 4B $0.05
- S Llama 3 8B Lunaris $0.05
- Ling 3.0 Flash $0.063
- Mistral Small 3 $0.08
- Llama 3.1 8B Instruct $0.08
- gpt-oss-20b $0.09
- N Nex-N2.5-Mini $0.10
As of October 21, 2026, inclusionAI Ling 3.0 Flash is the cheapest model at or above the median GPQA Diamond result, at $0.063 per million output tokens, ahead of Z.ai GLM 5.3 Flash ($0.50) and MiniMax M3 ($1.20).
This list narrows the best LLM for reasoning ranking.
| Rank | Model | Output $/Mtok | Coding evidence | GPQA Diamond | Coding index |
|---|---|---|---|---|---|
| 1 | Ling 3.0 Flash inclusionAI | $0.063 | 84.8% | Not benchmarked | |
| 2 | GLM 5.3 Flash Z.ai | $0.50 | 86.4% | 71.0 | |
| 3 | MiniMax M3 MiniMax | $1.20 | 92.7% | 57.8 | |
| 4 | GPT-5.6 Luna OpenAI | $1.20 | 91.7% | 61.5 | |
| 5 | MiniMax M2.7 MiniMax | $1.20 | 86.6% | 28.4 | |
| 6 | DeepSeek V4 Flash 0731 DeepSeek | $1.32 | 89.9% | Not benchmarked | |
| 7 | Qwen3.8 27B Qwen | $1.35 | 88.9% | 73.6 | |
| 8 | GLM 5.3 Z.ai | $1.40 | 88.1% | 73.1 | |
| 9 | $1.80 | 87.4% | Not benchmarked | ||
| 10 | DeepSeek V4 Pro 0423 DeepSeek | $1.9105 | 89.4% | 35.3 | |
| 11 | Qwen3.6 Plus Qwen | $1.95 | 87.4% | 37.4 | |
| 12 | DeepSeek V4 Pro 0813 DeepSeek | $1.98 | 92.4% | 54.6 | |
| 13 | Nemotron 3 Ultra NVIDIA | $2.40 | 86.1% | Not benchmarked | |
| 14 | Grok 4.3 xAI | $2.50 | 91.4% | 22.5 | |
| 15 | Grok 4.20 xAI | $2.50 | 88.6% | Not benchmarked | |
| 16 | Gemini 3 Flash Preview Google | $3.00 | 87.9% | 75.8 | |
| 17 | Kimi K2.6 Moonshot AI | $3.41 | 89.1% | 44.2 | |
| 18 | Gemini 3.7 Flash Google | $3.75 | 93.9% | 69.9 | |
| 19 | Gemini 3.6 Flash Google | $3.75 | 93.4% | 48.3 | |
| 20 | T Inkling Thinking Machines | $4.05 | 87.1% | 49.7 |
How this list is built
Generated from the dataset on every refresh. Nobody picks the order by hand.
Models whose GPQA Diamond result is at or above 84.8%, 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
- GPQA (Vals AI), fetched , board updated September 1, 2026