| # | Model | Input $/1M | Output $/1M | Monthly | vs cheapest |
|---|---|---|---|---|---|
| 1 | DeepSeek V4 Pro openDeepSeek | $0.435 | $0.87 | $6.09 | cheapest |
| 2 | Kimi K2.6 openMoonshot AI | $0.95 | $4 | $17.5 | ×2.9 vs cheapest |
| 3 | Qwen3.7 MaxAlibaba | $1.25 | $3.75 | $20.0 | ×3.3 vs cheapest |
| 4 | Gemini 3.5 FlashGoogle DeepMind | $1.50 | $9 | $33.0 | ×5.4 vs cheapest |
| 5 | Claude Sonnet 5Anthropic | $2 | $10 | $40.0 | ×6.6 vs cheapest |
| 6 | Gemini 3.1 ProGoogle DeepMind | $2 | $12 | $44.0 | ×7.2 vs cheapest |
| 7 | Claude Opus 4.8Anthropic | $5 | $25 | $100 | ×16 vs cheapest |
| 8 | GPT-5.5OpenAI | $5 | $30 | $110 | ×18 vs cheapest |
| 9 | Claude Fable 5Anthropic | $10 | $50 | $200 | ×33 vs cheapest |
how the math works
Monthly cost = input tokens × input price + output tokens × output price (both per 1M tokens). Coding is output-heavy — agents write diffs, tests and retries — so the output rate usually dominates a real bill even though input volume is larger. The table re-sorts live as you change the volumes.
These are list prices confirmed against each maker's pricing page (the same verified data behind our AI Coding Leaderboard). Prompt caching, batch APIs and committed-use discounts can cut real-world bills 50–90% for cache-friendly workloads — treat the numbers here as the ceiling, not the floor.
Raw pricing data: JSON · CSV. Models without a confirmed output price are excluded rather than estimated.