Compare Models

GPT-6 Astra

GPT-6 Astra is a model from OpenAI, released September 3, 2026. It ranks #6 of 44 models on the Vals Index with 63.13%. Its best result is #1 of 38 on Terminal-Bench Science.

Release Date: Sep 3, 2026

Developer OpenAI  🇺🇸
Context Window 1M
Max Output Tokens 128k
Token Costs (in/out) $10.00/50.00
Weights Private
Input Modalities

Accuracy

63.13 % ± 1.19

Cost / Test (Vals Index)

$ 18.46

Latency

27 min 48 s

Fallback Rate 0.00%
Refusal Rate 0.29%
Vals Index
BenchmarksAccuracyRankings
Vals Index

63.13%

±1.19
6/44
Vals RSI Index

27.07%

6/24
Code Migration

67.74%

±4.22
3/74
CUA-bench

19.17%

1/8
CyberBench v1.1

41.07%

±2.56
41/44
EMB

71.70%

±2.48
9/71
Finance Agent (v2)

53.54%

±2.08
28/75
Time Horizon Index: KSP

90.50%

±0.00
2/12
Legal Research Bench

39.42%

±3.40
27/74
MedCode

48.49%

±2.13
27/105
MedScribe

87.91%

±1.94
14/107
MysteryMechanism

53.15%

±3.36
1/24
ProofBench v1.1

99.00%

±1.00
5/47
SAGE

46.37%

±3.44
36/90
SRE Bench

56.87%

±3.07
1/32
Tax Agent Bench

63.34%

±3.15
24/66
Vibe Code Bench 1-100

27.64%

±4.08
4/21
Vibe Code Bench v1.1

89.59%

±2.17
6/109
BioMysteryBench

79.26%

±0.98
3/24
Harvey's Legal Agent Benchmark

5.42%

±1.17
31/75
IOI

100.00%

±0.00
2/41
ProgramBench

5.50%

±1.62
4/62
Terminal-Bench 4.0

59.60%

±4.40
3/44
Terminal-Bench Science

62.86%

±5.82
1/38
Proprietary BenchmarksAcademic BenchmarksIndustry Partners
Vals
Default Provider : OpenAI
Temperature: Default
Top P: Default
Top K: Default
Max Output Tokens: 128,000
Reasoning Effort: max

Compared with

GPT-6 Astra vs Claude Opus 5.5 24 shared benchmarks GPT-6 Astra vs Claude Sonnet 5.5 20 shared benchmarks GPT-6 Astra vs Claude Fable 5.1 23 shared benchmarks GPT-6 Astra vs Claude Fable 5 16 shared benchmarks GPT-6 Astra vs Claude Opus 5 24 shared benchmarks GPT-6 Astra vs Muse Spark 1.3 Max 19 shared benchmarks GPT-6 Astra vs Gemini 3.8 Flash 24 shared benchmarks GPT-6 Astra vs Claude Opus 4.8 16 shared benchmarks