GPT-6 Astra
Flagshipby OpenAI
GPT-6 Astra is the OpenAI flagship, released September 3, 2026 and pitched squarely at computer use: driving a browser, filling forms, editing spreadsheets and operating desktop applications the way a person would. API pricing is $10 per million input tokens and $50 per million output, with cached input at $1 and batch at half price. Two modifiers matter more than the headline rate. Requests above 272,000 input tokens bill on a separate long-context tier at $20 input, $2 cached and $75 output, which is easy to trip on exactly the long-horizon agent runs Astra is sold for, because an agent accumulating screenshots and tool output crosses that line without deciding to. Fast mode doubles every rate. Astra is also the first model OpenAI has placed in the Critical cybersecurity tier of its own Preparedness Framework, after pausing it in August: it scored 100 percent on ExploitBench and chained previously unknown vulnerabilities without step-by-step human direction. Rollout began as a limited preview for vetted cybersecurity partners before reaching ChatGPT Plus, Pro, Business and Enterprise, the API, and AWS.
Input Price
$10.00
per 1M tokens
Output Price
$50.00
per 1M tokens
Context Window
1M
tokens
Released
2026-09
API access
Capabilities
Key Strengths
- ✓72.6 percent on OSWorld 2.0 computer use, against 65.7 for GPT-5.6 Sol
- ✓96.0 on GPQA Diamond and 91.5 on BrowseComp
- ✓Roughly 40 minutes per evaluated task versus about 75 for Sol, which is what makes retrying an agent affordable
- ✓Cached input at $1 per 1M, a tenth of the standard rate
- ✓Strong long-context retrieval: 96.3 percent on MRCR v2 8-needle in the 512K to 1M band
Best For
- ▸Computer-use and browser-driving agents
- ▸Long-horizon workflows that accumulate screen and tool state
- ▸Frontier reasoning and research
- ▸Authorized security research under the OpenAI vetted-access program
Benchmark Scores
| Benchmark | Score | Description |
|---|---|---|
| GPQA Diamond | 96.0 | Graduate-level science questions verified by domain experts |
| BrowseComp | 91.5 | Agentic web search and browsing over hard-to-find facts |
Scores sourced from public benchmark datasets. See full benchmark leaderboard for all models.
Pricing Details
Input tokens
$10.00
per 1M tokens
Output tokens
$50.00
per 1M tokens
Estimated cost per 1K requests
$35.00
~1K input + ~500 output tokens avg
Prices are subject to change. Check the official documentation for current pricing. See the cost calculator for detailed estimates.
Related Models
GPT-5.6 Sol
FlagshipOpenAI
$4.00 in / $20.00 out
GPT-5.6 Terra
Mid-tierOpenAI
$2.00 in / $12.00 out
GPT-5.6 Luna
BudgetOpenAI
$0.20 in / $1.20 out
GPT-5.5
FlagshipOpenAI
$5.00 in / $30.00 out
GPT-4o
FlagshipOpenAI
$2.50 in / $10.00 out
GPT-4o-mini
BudgetOpenAI
$0.15 in / $0.60 out
o1
FlagshipOpenAI
$15.00 in / $60.00 out
o3-mini
Mid-tierOpenAI
$1.10 in / $4.40 out
Claude Opus 5
FlagshipAnthropic
$5.00 in / $25.00 out
Claude Fable 5.1
FlagshipAnthropic
$10.00 in / $50.00 out
Claude Fable 5
FlagshipAnthropic
$10.00 in / $50.00 out
Claude Opus 4.8
FlagshipAnthropic
$5.00 in / $25.00 out