Last verified on 2026-08-03. OpenAI currently documents six reasoning-effort values, not five: none, low, medium, high, xhigh, and max.
The one-minute answer
GPT-5.6 is a three-model family. The gpt-5.6 alias currently routes to the flagship Sol model. Terra balances capability and cost, while Luna targets cost-sensitive, high-volume workloads. All three support text and image input, text output, streaming, structured outputs, function calling, and the Responses, Chat Completions, and Batch APIs.
| Model | Official positioning | Good starting point for |
|---|---|---|
| GPT-5.6 Sol | Flagship for complex professional work | Hard coding, research, and complex agents |
| GPT-5.6 Terra | Balance of intelligence and cost | Everyday production, coding, content, analysis |
| GPT-5.6 Luna | Cost-sensitive, high-volume model | Classification, extraction, batch, light agents |
Context and output limits
All three currently have a 1,050,000-token context window, 922,000 maximum input, and 128,000 maximum output. Their knowledge cutoff is 2026-02-16. Long context is useful, but production systems should still retrieve, trim, and cache context deliberately.
Six reasoning levels
Set reasoning.effort to none, low, medium, high, xhigh, or max. The default is medium. Higher levels generally use more reasoning tokens and take longer. Start simple extraction at none or low and increase only when real evaluations justify it.
reasoning.mode is a separate choice: standard is the default, while pro spends more compute on difficult work and can increase latency and token usage. It does not use a separate model slug.
Standard pricing (USD per 1M tokens)
| Model | Input | Cached input | Cache write | Output |
|---|---|---|---|---|
| Sol | $5.00 | $0.50 | $6.25 | $30.00 |
| Terra | $2.00 | $0.20 | $2.50 | $12.00 |
| Luna | $0.20 | $0.02 | $0.25 | $1.20 |
For requests above 272K input tokens, long-context pricing applies to the full request: input is 2x standard and output is 1.5x standard. Batch and Flex table prices are 50% of standard; Fast short-context prices are 2x standard. Eligible regional data-residency endpoints may add 10%. Check the official pricing page and OpenTokenRouter live pricing before production settlement.
Choosing a model
- Start with Terra as a default and evaluate quality, latency, and cost on real tasks.
- Move to Sol only when difficult professional workloads show a measurable benefit.
- Evaluate Luna first for extraction, classification, and high-volume light work.
- Treat reasoning effort as a per-task control instead of forcing max everywhere.
Responses API example
curl "$OPEN_TOKEN_ROUTER_BASE_URL/v1/responses" \
-H "Authorization: Bearer $OPEN_TOKEN_ROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-5.6-terra",
"reasoning": {"effort": "medium"},
"input": "Design a pre-release checklist for this Go service"
}'Rate limits and quotas are not one permanent public number; they vary by account tier, project, and model. Check the provider dashboard Limits page instead of copying a single quota into evergreen documentation.