OpenAI’s GPT-5.6 family hit general availability this morning in three sizes — Luna, Terra, and Sol — with the flagship Sol priced at $5/$30 per million input/output tokens, matching Claude Opus on input but running $5 higher on output.
The three tiers, smallest to largest:
- Luna — $1/$6 per 1M input/output tokens
- Terra — $2.50/$15 per 1M input/output tokens
- Sol — $5/$30 per 1M input/output tokens
For context, Claude Opus sits at $5/$25 and Claude Fable 5 at $10/$50. But per-token pricing is a blunt instrument now — reasoning token counts vary between models on the same task, so actual spend depends on how efficiently a model thinks.
OpenAI’s headline claim targets long-running agentic work. On Agents’ Last Exam — an evaluation spanning 55 professional fields — GPT-5.6 Sol scored 53.6, beating Claude Fable 5’s adaptive reasoning by 13.1 points. At medium reasoning, Sol still outperforms Fable 5 by 11.4 points at roughly one-quarter the estimated cost.
| Model | Input/1M | Output/1M |
|---|---|---|
| GPT-5.6 Luna | $1 | $6 |
| GPT-5.6 Terra | $2.50 | $15 |
| GPT-5.6 Sol | $5 | $30 |
| Claude Opus | $5 | $25 |
| Claude Fable 5 | $10 | $50 |
NOTE
OpenAI says it “trained GPT-5.6 to get more useful work from every token” — meaning fewer reasoning tokens for equivalent output. Benchmark cost comparisons factor this in; real-world spend will depend on your specific workload mix.
Sol’s edge is efficiency, not raw price. It costs more per output token than Opus, but if it burns fewer reasoning tokens on agentic tasks, the total bill shrinks. Available now through the OpenAI API and Codex.