in the loop

GPT-6 Astra
#2 of 274 scored.

OpenAISep 3, 2026Closed

Capabilities Index

Range 163.3–171.5

166.4

Context

Takes file, image, text

1.05M

Input

Per 1M tokens

$10

Output

Per 1M tokens

$50

Benchmarks

How it scores.

  • GPQA Diamond94.4%

    1st of 186

  • FrontierMath93.7%

    1st of 81

  • ARC-AGI-295.0%

    1st of 79

  • Humanity's Last Exam52.5%

    1st of 40

  • SimpleQA Verified75.6%

    1st of 77

  • Mock AIME100%

    1st of 176

Everything else Epoch has for it

  • ProofBench99.0%
  • ARC-AGI98.5%
  • FrontierMath-Tier-4-v2-Private97.6%
  • DTBench95.5%
  • WeirdML93.6%
  • Mystery Game Puzzles82.4%
  • EBR-bench76.2%
  • LMCA75.7%
  • DeepSWE74.1%
  • Furniture Assembly71.4%
  • Chess Puzzles70.5%
  • Balrog68.3%
  • FrontierSWE65.5%
  • APEX-Agents64.7%
  • FrontierCode53.3%
  • MirrorCode46.7%
  • Remote Labor Index20.8%

On the index

Around it.

  1. #1Claude Opus 5.5Anthropic · $4.00 in167.3
  2. #2GPT-6 AstraOpenAI · $10 in166.4
  3. #3GPT-6.1 SolOpenAI · $2.00 in166.1
  4. #4Claude Sonnet 5.5Anthropic · $2.00 in165.0
  5. #5Claude Fable 5.1Anthropic · $10 in164.7

Sources