All prices are in USD per 1 million tokens. The input rate applies to tokens not found in the cache.
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT30% of official API rates | |||
| GPT-6.2-astra | $2.40 | $0.12 | $12.00 |
| GPT-6.2-sol | $0.60 | $0.03 | $3.00 |
| GPT-6.1-sol | $0.60 | $0.03 | $3.00 |
| GPT-6.1-terra | $0.30 | $0.015 | $1.50 |
| GPT-6-astra | $3.00 | $0.15 | $15.00 |
| GPT-6-luna | $0.03 | $0.003 | $0.15 |
| Claude30% of official API rates | |||
| Claude Fable 5.2 | $2.40 | $0.06 | $12.00 |
| Claude Opus 5.5 | $1.20 | $0.06 | $6.00 |
| Claude Sonnet 5.5 | $0.60 | $0.03 | $3.00 |
| Claude Haiku 5.5Input ≤ 100k tokens | $0.03 | $0.003 | $0.15 |
| Claude Haiku 5.5Input > 100k tokens | $0.15 | $0.015 | $0.75 |
| DeepSeek50% of official API rates | |||
| DeepSeek V4.1 FlashPeak | $0.15 | $0.003 | $0.60 |
| DeepSeek V4.1 FlashOff-peak | $0.075 | $0.0015 | $0.30 |
GPT prices shown are for standard context lengths. DeepSeek peak hours are Monday to Friday, 01:00–04:00 and 06:00–10:00 UTC, except on Chinese public holidays. Off-peak rates apply at all other times.
How it works
01
Model choice for research tasks
Use different models in the same project to assess literature, modify code, and diagnose experiment results.
- Long-context research
- Complex code and debugging
- Multi-round experiment analysis
02
Plan credits
Model calls and compute rentals draw from the same balance based on actual usage.
- $19 per month
- 60 credits
- Models and compute
03
Spend the budget on more iterations
Research value usually comes from repeated testing and correction. Lower model cost gives the agent room for more useful loops.
- Continuous execution
- Repeated validation
- Reproducible results