Selection guide
Which OpenAI model fits the workload?
Start with workload volume and failure cost. Luna targets high-volume work, Terra balances cost and capability, Sol reserves the family’s highest-capability tier for difficult tasks, and GPT-4.1 remains a separately documented long-context API model.
API + app
GPT-5.6 Luna
Consider for: High-volume agent and production tasks where listed token cost is the first constraint.
Account for: Lower reported capability ceiling than Terra and Sol.
API + app
GPT-5.6 Terra
Consider for: General production work that needs a middle ground between family capability and token price.
Account for: Costs more than Luna and does not reach the reported Sol tier.
API + app
GPT-5.6 Sol
Consider for: Difficult reasoning, coding, science, and long-horizon agent work where failure cost can justify premium pricing.
Account for: Highest listed input and output prices in the documented OpenAI set.
API
GPT-4.1
Consider for: Long-input code and document analysis that specifically targets the GPT-4.1 API record.
Account for: Older benchmark coverage is not directly comparable with the GPT-5.6 launch taxonomy.
Editorial OpenAI comparisons
- GPT-5.6 Sol vs GPT-5.6 Terra — Both share the GPT-5.6 tool stack; Sol buys a higher capability ceiling while Terra lowers token costs by half.
- GPT-5.6 Terra vs GPT-5.6 Luna — Terra is the balanced tier; Luna cuts input and output list prices while giving up some peak reasoning performance.
OpenAI model specifications and prices
| Model | Access | Released | Context | Input / 1M | Output / 1M |
|---|---|---|---|---|---|
| GPT-5.6 Luna | API + app | July 9, 2026 | 1.0M | $1.00 | $6.00 |
| GPT-5.6 Terra | API + app | July 9, 2026 | 1.0M | $2.50 | $15.00 |
| GPT-5.6 Sol | API + app | July 9, 2026 | 1.0M | $5.00 | $30.00 |
| GPT-4.1 | API | April 14, 2025 | 1.0M | $2.00 | $8.00 |
Prices are provider-listed USD snapshots per million tokens where represented. Caching, batch processing, long-context rules, modalities, regions, and service tiers can change effective cost.
Reported benchmark coverage
| Model | MMLU-Pro | GPQA Diamond | SWE-bench Verified | LiveCodeBench | SWE-Bench Pro | Artificial Analysis Intelligence Index |
|---|---|---|---|---|---|---|
| GPT-5.6 Luna | — | 92.3 | — | — | 62.7 | 51.2 |
| GPT-5.6 Terra | — | 92.9 | — | — | 63.4 | 55 |
| GPT-5.6 Sol | — | 94.6 | — | — | 64.6 | 58.9 |
| GPT-4.1 | 83.7 | 66.3 | 54.6 | 44.7 | — | — |
A dash means the exact evaluation was not retained for that model. Unlike benchmark families and source settings are not merged into an overall score.
Provider and model sources
Each model keeps its own source record and limitations. Review the provider API reference for current authentication, quotas, SDK status, and pricing rules before implementation.