Cost · best for
Best AI model for Cheap Bulk Inference (2026)
Lowest cost-per-million for high-volume jobs. Ranked from 346 live models on the OpenRouter catalog, weighted for low cost, low latency.
| # | Model | Score | In / 1M | Out / 1M | Context | |
|---|---|---|---|---|---|---|
| 1 | Pareto Code Routeropenrouter/pareto-code | 1000138 | $-1000000.00 | $-1000000.00 | 200,000 | Try → |
| 2 | Body Builder (beta)openrouter/bodybuilder | 1000138 | $-1000000.00 | $-1000000.00 | 128,000 | Try → |
| 3 | Auto Routeropenrouter/auto | 1000138 | $-1000000.00 | $-1000000.00 | 2,000,000 | Try → |
| 4 | Google: Gemma 4 26B A4B (free)google/gemma-4-26b-a4b-it:free | 138 | Free | Free | 262,144 | Try → |
| 5 | Google: Gemma 4 31B (free)google/gemma-4-31b-it:free | 138 | Free | Free | 262,144 | Try → |
| 6 | Qwen: Qwen3.5-9Bqwen/qwen3.5-9b | 137 | $0.10 | $0.15 | 262,144 | Try → |
| 7 | Google: Gemma 4 26B A4B google/gemma-4-26b-a4b-it | 137 | $0.07 | $0.35 | 262,144 | Try → |
| 8 | ByteDance Seed: Seed-2.0-Minibytedance-seed/seed-2.0-mini | 137 | $0.10 | $0.40 | 262,144 | Try → |
| 9 | Qwen: Qwen3.5-Flashqwen/qwen3.5-flash-02-23 | 137 | $0.07 | $0.26 | 1,000,000 | Try → |
| 10 | ByteDance Seed: Seed 1.6 Flashbytedance-seed/seed-1.6-flash | 137 | $0.07 | $0.30 | 262,144 | Try → |
| 11 | Google: Gemini 2.5 Flash Lite Preview 09-2025google/gemini-2.5-flash-lite-preview-09-2025 | 137 | $0.10 | $0.40 | 1,048,576 | Try → |
| 12 | OpenAI: GPT-5 Nanoopenai/gpt-5-nano | 137 | $0.05 | $0.40 | 400,000 | Try → |
| 13 | Google: Gemini 2.5 Flash Litegoogle/gemini-2.5-flash-lite | 137 | $0.10 | $0.40 | 1,048,576 | Try → |
| 14 | OpenAI: GPT-4.1 Nanoopenai/gpt-4.1-nano | 137 | $0.10 | $0.40 | 1,047,576 | Try → |
| 15 | Google: Gemini 2.0 Flash Litegoogle/gemini-2.0-flash-lite-001 | 137 | $0.07 | $0.30 | 1,048,576 | Try → |
How we ranked these
For Cheap Bulk Inference, we weight models on low cost, low latency. Higher means better. Scores combine OpenRouter's model metadata (context length, modality support, tool calling, structured output, reasoning capability) with public pricing. See full methodology →