Quick answer
Start with Sol for complex coding and reasoning. Escalate to Astra for maximum capability; use Luna for repeatable high-volume work.
- Released
- September 3, 2026
- API model
- gpt-6-sol
- Price
- $2 in / $10 out per MTok; $0.20 cached input; long-context rates apply above 272K input tokens
- Context
- 1,050,000 tokens (128K max output)
01
GPT-6 Sol: verified facts
| Field | Verified value |
|---|---|
| Vendor | OpenAI |
| Release | September 3, 2026 |
| API identifier | gpt-6-sol |
| Input / output price | $2 in / $10 out per MTok |
| Cache / fast pricing | $0.20 cached input; long-context rates apply above 272K input tokens |
| Context / max output | 1,050,000 tokens (128K max output) |
| Availability | GPT-6 everyday driver for complex coding and agentic workflows |
Every number on this page was checked against the linked vendor documentation on September 23, 2026. Prices are API list prices per million tokens; product subscriptions and regional availability are separate.
02
Where this model fits
OpenAI recommends Sol for demanding reasoning and everyday complex coding. It supports none through max reasoning effort and shares Astra's 1.05M-token context and 128K maximum output at one fifth of Astra's base token price.
Your AI workspace for shipping software.
Download AIDEN free and point it at your existing Claude Code or Codex setup. No credit card, running in minutes.
Download AIDEN freeFree to start · macOS 12+ · No credit card required
03
When to use it
Use case 1
Default Codex model for serious implementation work
Use case 2
Agentic workflows that need judgment and completeness
Use case 3
High-volume engineering where Astra would be too expensive
| Strengths | Trade-offs |
|---|---|
| One fifth of Astra's base token price | Less capable than Astra on OpenAI's model-selection ladder |
| 1.05M-token context and 128K maximum output | Chat Completions function calling requires reasoning effort none |
| Supports reasoning from none through max in the Responses API | Requests above 272K input tokens use higher long-context rates |
04
Published benchmarks
05