Agent-ready GPT for coding and computer-use workflows at a lower cost
Capabilities
ReasoningYes
Tool callingYes
Structured outputYes
TemperatureFixed
AttachmentYes
Open weightsNo
Limits
Context1,050,000 tokens
Max input922,000 tokens
Max output128,000 tokens
Details
Familygpt
Knowledge cutoff2025-08-31
Release date2026-03-05
Last updated2026-03-05
Modalities
Inputtext, image, pdf
Outputtext
Benchmarks
SWE-Bench Pro
59.1
resolve rate
SWE-Atlas Codebase QnA
40.8
score
SWE-Atlas Codebase QnA
36.3
score
SWE-Atlas Refactoring
44.29
score
SWE-Atlas Test Writing
44.36
score
SWE-Atlas Test Writing
40
score
Artificial Analysis Coding Agent Index
53.6
average pass@1
SWE-Atlas Codebase QnA
72.4
pass@1
Artificial Analysis Coding Agent Index
52.2
average pass@1
SWE-Atlas Codebase QnA
72.9
pass@1
Comments (0)
Sign in to leave a comment
No comments yet.