| # | Model | Per month | Score |
|---|---|---|---|
| 1 | Granite 4.0 MicroIBM | $0.73 | 25.0% |
| 2 | gpt-oss-20bOpenAI | $0.87 | 51.2% |
| 3 | Nova Micro 1.0Amazon | $1.05 | 22.5% |
| 4 | Phi 4Microsoft | $1.40 | 41.2% |
| 5 | Ministral 3 3B 2512Mistral | $1.50 | 26.2% |
| 6 | Ling 3.0 Flash VLinclusionAI | $1.50 | — |
| 7 | Mercury 2.5Inception | $1.53 | 58.2% |
| 8 | Granite 4.2 8BIBM | $2.19 | 36.2% |
| 9 | Ministral 3 8B 2512Mistral | $2.25 | 42.5% |
| 10 | Llama 4 ScoutMeta | $2.50 | 41.2% |
| 11 | Llama 3.3 70B InstructMeta | $2.60 | 42.5% |
| 12 | Ministral 3 14B 2512Mistral | $3.00 | 33.8% |
| 13 | Llama 4 MaverickMeta | $5.14 | 59.0% |
| 14 | Qwen3 Coder NextQwen | $5.20 | 62.5% |
| 15 | GPT-6 LunaOpenAI | $5.23 | 72.5% |
| 16 | Claude Haiku 5.5Anthropic | $5.39 | 68.8% |
| 17 | LFM2.5-2.6BLiquidAI | $6.00 | — |
| 18 | Nemotron 3.5 LightningNVIDIA | $6.21 | 43.8% |
| 19 | Ling 3.1 FlashinclusionAI | $6.39 | 41.2% |
| 20 | MiMo-V2.6-FlashXiaomi | $6.39 | 62.5% |
| 21 | GLM 5.3 FlashZ.ai | $6.44 | 67.5% |
| 22 | Solar Mini 4Upstage | $6.48 | 41.2% |
| 23 | gpt-oss-120bOpenAI | $6.72 | 58.8% |
| 24 | Qwen3 Next 80B A3B ThinkingQwen | $7.50 | — |
| 25 | MiniMax M3.1 FlashMiniMax | $9.30 | 66.2% |
| 26 | Reka Flash 3rekaai | $10.06 | 1.2% |
| 27 | DeepSeek V4.1 Flash 1017DeepSeek | $11.73 | 73.8% |
| 28 | Inkling SmallThinking Machines | $11.74 | 67.5% |
| 29 | Gemma 4 26B A4BGoogle | $12.58 | 15.0% |
| 30 | DeepSeek V4 Flash Vision ExpDeepSeek | $13.52 | 71.2% |
| 31 | GLM 5.3Z.ai | $13.87 | 67.5% |
| 32 | Gemma 4 31BGoogle | $15.58 | 30.0% |
| 33 | Hy3Tencent | $16.06 | 58.8% |
| 34 | Nemotron 3.5 Super VLNVIDIA | $17.57 | 40.0% |
| 35 | Qwen3.5-122B-A10BQwen | $20.00 | — |
| 36 | Solar Pro 4Upstage | $21.49 | 45.0% |
| 37 | Mistral Small 4Mistral | $22.04 | 32.5% |
| 38 | MiMo-V2.6-ProXiaomi | $22.70 | 68.8% |
| 39 | Hermes 4 405BNous | $25.00 | 26.2% |
| 40 | Trinity Large ThinkingArcee AI | $30.65 | 28.8% |
| 41 | MiniMax M3MiniMax | $32.29 | 57.5% |
| 42 | Qwen3.6 35B A3BQwen | $34.77 | 37.5% |
| 43 | Qwen3.7 PlusQwen | $37.66 | 60.0% |
| 44 | InklingThinking Machines | $38.82 | 66.2% |
| 45 | Gemini 3.5 Flash LiteGoogle | $39.13 | 75.0% |
| 46 | LongCat 2.0Meituan | $39.21 | 48.8% |
| 47 | GPT-6.1 TerraOpenAI | $40.25 | 73.8% |
| 48 | GPT-6.1 Terra ProOpenAI | $47.87 | 76.2% |
| 49 | Step 5 PreviewStepFun | $51.44 | 63.8% |
| 50 | Qwen3.8 27BQwen | $52.50 | 63.8% |
| 51 | Gemini 3.8 FlashGoogle | $57.74 | 82.5% |
| 52 | Muse Spark OpenMeta | $64.59 | 77.5% |
| 53 | Nemotron 3 UltraNVIDIA | $64.80 | 60.0% |
| 54 | Claude Sonnet 5.5Anthropic | $70.00 | 78.8% |
| 55 | Kimi K2.7 CodeMoonshotAI | $71.82 | 65.0% |
| 56 | Gemini 4 ArgonGoogle | $72.97 | 84.8% |
| 57 | Command ACohere | $75.00 | 45.0% |
| 58 | GPT-6.1 SolOpenAI | $78.93 | 76.2% |
| 59 | GPT-6 SolOpenAI | $81.19 | 73.8% |
| 60 | DeepSeek V4 Pro 0813DeepSeek | $82.47 | 71.2% |
| 61 | Nova Premier 1.0Amazon | $87.50 | — |
| 62 | GPT-5.6 TerraOpenAI | $123 | 75.0% |
| 63 | Qwen3.8 Max (0902)Qwen | $133 | 75.0% |
| 64 | o3OpenAI | $134 | 72.5% |
| 65 | Mistral Large 4Mistral | $140 | 50.0% |
| 66 | Kimi K3.1 ProMoonshotAI | $154 | 77.5% |
| 67 | GPT-5.3-CodexOpenAI | $155 | 82.5% |
| 68 | Qwen3.5 397B A17BQwen | $160 | 32.5% |
| 69 | GPT-5.4OpenAI | $168 | 72.5% |
| 70 | Claude Opus 5.5Anthropic | $178 | 85.0% |
| 71 | Qwen3.8 2.4T A95BQwen | $184 | 57.5% |
| 72 | Gemini 3.1 Pro PreviewGoogle | $196 | 77.5% |
| 73 | Grok 4.7SpaceXAI | $276 | 80.0% |
| 74 | Mistral Medium 3.5Mistral | $287 | 28.8% |
| 75 | Muse WonderMeta | $334 | 78.8% |
| 76 | Claude Fable 5.1Anthropic | $350 | 81.2% |
| 77 | GPT-6 AstraOpenAI | $384 | 76.2% |
| 78 | GPT-5.5 ProOpenAI | $1,200 | — |
Tools · prices from OpenRouter, 27 Oct 2026
LLM API price calculator
Enter your expected usage to estimate the monthly bill for each of the 78 models we test. Reasoning models also bill their hidden thinking; turn that on to add the reasoning tokens each model used per question in our runs.
Frequently asked questions
- How is the monthly cost calculated?
- Requests per month × (input tokens × input price + output tokens × output price), using each model's list price per million tokens from OpenRouter's catalogue.
- Why add reasoning tokens?
- Reasoning models bill their thinking as output tokens even though it never appears in the answer. On our benchmark questions 58 models used reasoning tokens; Reka Flash 3 used the most, 4,030 per question on average. Your prompts may need more or fewer.
- Which model was cheapest in your tests?
- Granite 4.0 Micro by IBM had the lowest billed cost per benchmark question: $0.000051 on average, reasoning tokens included.