GPT 6 Astra

OpenAI's most capable model for the hardest end-to-end work, with state-of-the-art performance in computer use, browsing, software engineering, science, and professional tasks. Astra is designed to carry long, multistep workflows across code, browsers, and professional software through to a finished result.

Reasoning
Vision
Native PDF input
Tool Calling
Structured Output

Added Sep 3, 2026

Pricing

Auto routing · per 1M tokens

Input: $10.00
Output: $50.00
Cache read: $1.00

Compare provider prices

Specifications

Context window: 1.1M
Max output: 128K
Avg output (7d): 518 tokens; Longer than 46% of models

Benchmarks

Artificial Analysis

Vectara

Design Arena

Sourced from Artificial Analysis.

Intelligence Index

49.6

Better than 97% of models compared

Coding Index

76.7

Better than 96% of models compared

Agentic Index

46.0

Better than 87% of models compared

Agentic work

AutomationBench-AA

Workflow automation with guardrail penalties

64.6%

Better than 93% of models compared

AA-Briefcase

Agentic knowledge work (Elo)

1459 Elo

Better than 82% of models compared

GDPval-AA v2

Economically valuable tasks (Elo)

1468 Elo

Better than 82% of models compared

Document reasoning

GDP.pdf

Professional PDF reasoning: all-pass rate

30.4%

Better than 96% of models compared

AA-LCR v1.1

Long context reasoning with updated grading

79.7%

Better than 85% of models compared

Reasoning

HLE

Humanity's Last Exam

52.7%

Better than 97% of models compared

CritPt

Research-level physics reasoning

29.1%

Coding

Terminal-Bench v4.0

Practical coding and terminal tasks

49.5%

Better than 92% of models compared

SciCode

Python programming for scientific computing

54.2%

Better than 72% of models compared

Knowledge

AA-Omniscience Accuracy

Proportion of correctly answered questions

60.6%

AA-Omniscience Hallucination Rate

Rate of incorrect answers among non-correct responses

46.5%

Legacy benchmarks

GPQA Diamond (legacy)

Graduate-level scientific reasoning

93.9%

Better than 98% of models compared

AA-LCR (unversioned / legacy)

Long context reasoning evaluation

79.7%

Better than 84% of models compared

GDPval-AA (unversioned / legacy)

Economically valuable tasks

49.3%

Last updated Oct 14, 2026