S1哥 今日赠品 · 2026-10-31
配套教程：https://x.com/swordBluesy/status/2116506486272844138
Published API list prices, USD per 1M tokens, as reported Oct 29-30, 2026 (check each lab's pricing page before you build). | Model | Input | Output | 3:1 blended (uncached) | |---|---|---|---| | Kimi K3.1 Pro (Moonshot) | $2.00 | $10.00 | $4.00 | | MiniMax M3.1 Flash | $0.15 | $0.60 | $0.2625 | | Qwen3.8-Max (Alibaba, Singapore region) | $2.00 | $6.00 | $3.00 | | Qwen3.7-Max | $2.50 | $7.50 | $3.75 | | DeepSeek V4 Flash 0731 (OpenRouter listing) | $0.04 | $0.30 | about $0.09 | Notes - Kimi K3.1 Pro is about 15x MiniMax M3.1 Flash on blended price. - One 43-question run (OrcaRouter, Oct 26): MiniMax 41 correct for $4.09, Kimi 42 correct for $13.47. - Artificial Analysis index v5 (private questions, Oct 27): Kimi K3.1 Pro 46, DeepSeek V4.1 Flash 32, MiniMax M3.1 Flash 29. Claude Opus 5.5 58, GPT-6 Astra 57. - Moonshot's K2 series closes to new signups Oct 31 and retires end of December. Copy-paste prompt: "Here is my task and 20 sample inputs. Run them through the cheap model and the expensive model, and tell me the accuracy gap and the cost gap. Recommend which to ship."