GPT-6.2 Astra
Language Available Full comparison ↗ openai/gpt-6.2-astra · by OpenAI
· decoder-only-transformer
Pricing — 1 offering(s)
About this offering: OpenAI's GPT-6.2 Astra, the top tier of the GPT-6 family, launched 2026-11-17 at $8 input and $40 output per 1M tokens, 20% below GPT-6 Astra's $10 and $50. OpenAI now recommends it as the place to start for complex reasoning and coding. Above 272K input tokens, input and cache rates double and output costs 1.5 times as much, for the whole request: long-context cache writes are $20 (30-minute) and $40 (12-hour) per 1M tokens. Batch costs half the Standard rate at every context length. OpenAI also lists a legal configuration, gpt-6.2-astra-law, at the same prices.
Input tokens
- $8.00 / 1M tokens (input)
Current
2026-11-17 → present
Standard processing, short context (272K input tokens or fewer). Fast mode costs twice the Standard rate…
Standard processing, short context (272K input tokens or fewer). Fast mode costs twice the Standard rate; Ultrafast costs 3 times the Standard rate through February 28, 2027 and 6 times from March 1, 2027 (Pacific time).
Output tokens
- $40.00 / 1M tokens (output)
Current
2026-11-17 → present
Standard processing, short context (272K input tokens or fewer).
Cached input tokens
- $0.40 / 1M tokens (cached input) Current 2026-11-17 → present
Cache writes (30-minute retention)
- $10.00 / 1M tokens (cache write) Current 2026-11-17 → present
Cache writes (12-hour retention)
- $20.00 / 1M tokens (cache write) Current 2026-11-17 → present
Input tokens (prompts over 272K)
- $16.00 / 1M tokens (input)
Current
2026-11-17 → present
Applies to the full request when it has more than 272K input tokens.
Output tokens (prompts over 272K)
- $60.00 / 1M tokens (output) Current 2026-11-17 → present
Cached input tokens (prompts over 272K)
- $0.80 / 1M tokens (cached input) Current 2026-11-17 → present
Batch
Batch input tokens
- $4.00 / 1M tokens (input) Current 2026-11-17 → present
Batch output tokens
- $20.00 / 1M tokens (output) Current 2026-11-17 → present
Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.
See how GPT-6.2 Astra fits into a cost-aware routing setup
See how →Capability profile
How GPT-6.2 Astra rates across core capability dimensions, with the task-level evidence behind each rating.
1,050,000-token context window (922,000 max input), 128,000 max output tokens (developers.openai.com/api/docs/models/gpt-6.2-astra). Above 272K input tokens the whole request is billed at long-context rates.
OpenAI API docs: in the Responses API it supports web search, file search, image generation, code interpreter, hosted shell, apply patch, skills, computer use, MCP and tool search, plus the Multi-agent beta for delegating to subagents. Function calling and structured outputs are supported.
Top tier of the GPT-6 family ($8/$40 per 1M tokens, Standard, short context): 20% below GPT-6 Astra's $10/$50 and four times GPT-6.2 Sol. Cached input $0.40/1M; Batch is half price.
Ratings are estimated — limited independent data is available for this model.
Coding benchmarks
View the full coding benchmark leaderboard →| Benchmark | Score | Harness | Source |
|---|---|---|---|
| Terminal-Bench 4.0 | 63.1% 2026-11 | mini-swe-agent | 📊 leaderboard ↗ |
| Vibe Code Bench ↗ | 89.9% 2026-11 | openhands | 📊 leaderboard ↗ |
Operator guidance
Top tier of GPT-6, launched 2026-11-17 as the replacement for GPT-6 Astra at 20% lower prices ($8/$40 vs $10/$50 per 1M tokens). OpenAI's model overview names it the place to start for complex reasoning and coding. Step down to GPT-6.2 Sol (a quarter of the price) for balanced work, or GPT-6 Luna for cost-sensitive, high-volume workloads. Keep prompts under 272K input tokens where you can: above that, input and cache rates double and output costs 1.5 times as much. Reasoning effort none is rejected.
Use cases
- Complex reasoning, coding, computer use and document work, per OpenAI's model docs
- OpenAI's recommended starting point for its hardest end-to-end tasks
Limitations
- Reasoning, coding, instruction-following, multilingual and speed ratings are left out. No independent benchmark was sourced for this entry, and none is inferred; OpenAI's launch claims are vendor claims, not cited benchmarks.
- Capability ratings are qualitative, not a single cited benchmark run