readable AI benchmarks (simplified)

readable AI benchmarks

OpenAI

GPT-6 Astra (high)

reasoning · closed weights · Sep 3, 2026

GPT-6 Astra (high) has a Quality Score of 77.7, ranking 8th among 284 scored models. Per 1M tokens, pricing is $10.00 input, $50.00 output, $1.00 cached input, and $12.50 cached output. It accepts text and image input, outputs text, and has a 1M token context window. Its Reliability Score is 71.9, ranking 1st among 284 scored models, above average (total average: 35.4). Its Value Score is 43.5, high compared with the total Value average of 29.1.

Quality Score77.7coverage 94.7%: missing CursorBench 3.2 and DeepSWE v1.1
Value Score43.5
Reliability71.9
Cache Discount90%

Model specification

Reasoning
reasoning
Input modalities
text, image
Output modalities
text
Context window
1M tokens
Weights
closed weights

Token prices USD per 1M tokens

Input
$10.00
Output
$50.00
Cached Input
$1.00
Cached Output
$12.50
Cost per task
$0.958

Capability

Intelligence
60.3
Coding
77.1
Agentic
49.2
Omniscience
43.7
Correct
61.1%
Blended price
$7.70

Answer outcomes

Attempt rate: 78.5%

Correct 61.1%Incorrect 17.4%Abstained 21.5%

Artificial Analysis benchmarks

GDPval-AA v2
54%
τ³-Banking
40%
Terminal-Bench v2.1
90%
SciCode
52%
Humanity’s Last Exam
53%
GPQA Diamond
95%
CritPt
29%
AA-Omniscience
72%
AA-LCR
76%

Similar models