Data article
The cheapest capable AI models — October 2026.
Cheap is easy; cheap and capable is the trick. These are the lowest-cost models whose League Score sits at or above the field median.
Editor’s intro: this table regenerates from the model database every month. The pattern to watch: the capable-cheap tier is where prices fall fastest, because it is where makers compete for volume.
| # | Model | Maker | League Score | $/1M in | $/1M out | Blended |
|---|---|---|---|---|---|---|
| 1 | Claude Haiku 5.5 | Anthropic | 68 | $0.10 | $0.50 | $0.20 |
| 2 | DeepSeek V4.1 Flash | DeepSeek | 75 | $0.30 | $1.20 | $0.52 |
| 3 | Gemini 3.5 Flash-Lite | Google DeepMind | 70 | $0.30 | $2.50 | $0.85 |
| 4 | Gemini 3.8 Flash | Google DeepMind | 63 | $0.75 | $3.75 | $1.50 |
| 5 | Grok 4.3 | xAI (SpaceXAI) | 74 | $1.25 | $2.50 | $1.56 |
| 6 | Kimi K2.6 | Moonshot AI | 70 | $0.95 | $4.00 | $1.71 |
| 7 | DeepSeek V4 Pro (0813) | DeepSeek | 63 | $1.32 | $3.96 | $1.98 |
| 8 | Muse Spark 1.3 | Meta | 69 | $1.25 | $4.25 | $2.00 |
| 9 | Grok 4.7 | xAI (SpaceXAI) | 67 | $2.00 | $6.00 | $3.00 |
| 10 | GPT-6.1 Sol | OpenAI | 94 | $2.00 | $10.00 | $4.00 |
Run your own workload through the cost calculator — the cheapest model for a million tokens is not always the cheapest for your mix of input and output.