wehub-resource-sync
216 lines
4.6 KiB
TOML
216 lines
4.6 KiB
TOML
name = "GPT-5.4"
|
|
description = "Agent-ready GPT for coding and computer-use workflows at a lower cost"
|
|
family = "gpt"
|
|
release_date = "2026-03-05"
|
|
last_updated = "2026-03-05"
|
|
attachment = true
|
|
reasoning = true
|
|
temperature = false
|
|
tool_call = true
|
|
structured_output = true
|
|
knowledge = "2025-08-31"
|
|
open_weights = false
|
|
|
|
[limit]
|
|
context = 1_050_000
|
|
input = 922_000
|
|
output = 128_000
|
|
|
|
[modalities]
|
|
input = ["text", "image", "pdf"]
|
|
output = ["text"]
|
|
|
|
[[benchmarks]]
|
|
name = "SWE-Bench Pro"
|
|
score = 59.1
|
|
metric = "resolve rate"
|
|
dataset = "public"
|
|
source = "https://labs.scale.com/leaderboard/swe_bench_pro_public"
|
|
|
|
[[benchmarks]]
|
|
name = "SWE-Atlas Codebase QnA"
|
|
score = 40.8
|
|
metric = "score"
|
|
harness = "Codex"
|
|
source = "https://labs.scale.com/leaderboard/sweatlas-qna"
|
|
|
|
[[benchmarks]]
|
|
name = "SWE-Atlas Codebase QnA"
|
|
score = 36.3
|
|
metric = "score"
|
|
harness = "Mini-SWE-Agent"
|
|
source = "https://labs.scale.com/leaderboard/sweatlas-qna"
|
|
|
|
[[benchmarks]]
|
|
name = "SWE-Atlas Refactoring"
|
|
score = 44.29
|
|
metric = "score"
|
|
harness = "Codex"
|
|
source = "https://labs.scale.com/leaderboard/sweatlas-refactoring"
|
|
|
|
[[benchmarks]]
|
|
name = "SWE-Atlas Test Writing"
|
|
score = 44.36
|
|
metric = "score"
|
|
harness = "Codex CLI"
|
|
source = "https://labs.scale.com/leaderboard/sweatlas-tw"
|
|
|
|
[[benchmarks]]
|
|
name = "SWE-Atlas Test Writing"
|
|
score = 40
|
|
metric = "score"
|
|
harness = "Mini-SWE-Agent"
|
|
source = "https://labs.scale.com/leaderboard/sweatlas-tw"
|
|
|
|
[[benchmarks]]
|
|
name = "Artificial Analysis Coding Agent Index"
|
|
score = 53.6
|
|
metric = "average pass@1"
|
|
harness = "Codex"
|
|
variant = "medium"
|
|
source = "https://artificialanalysis.ai/agents/coding-agents"
|
|
|
|
[[benchmarks]]
|
|
name = "SWE-Atlas Codebase QnA"
|
|
score = 72.4
|
|
metric = "pass@1"
|
|
harness = "Codex"
|
|
variant = "medium"
|
|
source = "https://artificialanalysis.ai/agents/coding-agents"
|
|
|
|
[[benchmarks]]
|
|
name = "SWE-Bench Pro"
|
|
score = 18.4
|
|
metric = "pass@1"
|
|
harness = "Codex"
|
|
variant = "medium"
|
|
dataset = "hard-aa"
|
|
source = "https://artificialanalysis.ai/agents/coding-agents"
|
|
|
|
[[benchmarks]]
|
|
name = "Terminal-Bench"
|
|
score = 69.8
|
|
metric = "pass@1"
|
|
harness = "Codex"
|
|
variant = "medium"
|
|
version = "2.1"
|
|
source = "https://artificialanalysis.ai/agents/coding-agents"
|
|
|
|
[[benchmarks]]
|
|
name = "Artificial Analysis Coding Agent Index"
|
|
score = 52.2
|
|
metric = "average pass@1"
|
|
harness = "Cursor CLI"
|
|
variant = "medium"
|
|
source = "https://artificialanalysis.ai/agents/coding-agents"
|
|
|
|
[[benchmarks]]
|
|
name = "SWE-Atlas Codebase QnA"
|
|
score = 72.9
|
|
metric = "pass@1"
|
|
harness = "Cursor CLI"
|
|
variant = "medium"
|
|
source = "https://artificialanalysis.ai/agents/coding-agents"
|
|
|
|
[[benchmarks]]
|
|
name = "SWE-Bench Pro"
|
|
score = 18.9
|
|
metric = "pass@1"
|
|
harness = "Cursor CLI"
|
|
variant = "medium"
|
|
dataset = "hard-aa"
|
|
source = "https://artificialanalysis.ai/agents/coding-agents"
|
|
|
|
[[benchmarks]]
|
|
name = "Terminal-Bench"
|
|
score = 64.7
|
|
metric = "pass@1"
|
|
harness = "Cursor CLI"
|
|
variant = "medium"
|
|
version = "2.1"
|
|
source = "https://artificialanalysis.ai/agents/coding-agents"
|
|
|
|
[[benchmarks]]
|
|
name = "Terminal-Bench"
|
|
score = 75.1
|
|
metric = "success rate"
|
|
version = "2.0"
|
|
source = "https://openai.com/index/introducing-gpt-5-5/"
|
|
date = "2026-04-23"
|
|
|
|
[[benchmarks]]
|
|
name = "GPQA Diamond"
|
|
score = 92.8
|
|
metric = "accuracy"
|
|
source = "https://openai.com/index/introducing-gpt-5-5/"
|
|
date = "2026-04-23"
|
|
|
|
[[benchmarks]]
|
|
name = "Humanity's Last Exam"
|
|
score = 39.8
|
|
metric = "accuracy"
|
|
variant = "no tools"
|
|
source = "https://openai.com/index/introducing-gpt-5-5/"
|
|
date = "2026-04-23"
|
|
|
|
[[benchmarks]]
|
|
name = "Humanity's Last Exam"
|
|
score = 52.1
|
|
metric = "accuracy"
|
|
variant = "with tools"
|
|
source = "https://openai.com/index/introducing-gpt-5-5/"
|
|
date = "2026-04-23"
|
|
|
|
[[benchmarks]]
|
|
name = "OSWorld-Verified"
|
|
score = 75.0
|
|
metric = "success rate"
|
|
source = "https://openai.com/index/introducing-gpt-5-5/"
|
|
date = "2026-04-23"
|
|
|
|
[[benchmarks]]
|
|
name = "BrowseComp"
|
|
score = 82.7
|
|
metric = "accuracy"
|
|
source = "https://openai.com/index/introducing-gpt-5-5/"
|
|
date = "2026-04-23"
|
|
|
|
[[benchmarks]]
|
|
name = "GDPval"
|
|
score = 83.0
|
|
metric = "wins or ties"
|
|
source = "https://openai.com/index/introducing-gpt-5-5/"
|
|
date = "2026-04-23"
|
|
|
|
[[benchmarks]]
|
|
name = "ARC-AGI-2"
|
|
score = 73.3
|
|
metric = "accuracy"
|
|
variant = "Verified"
|
|
source = "https://openai.com/index/introducing-gpt-5-5/"
|
|
date = "2026-04-23"
|
|
|
|
[[benchmarks]]
|
|
name = "FrontierMath"
|
|
score = 47.6
|
|
metric = "accuracy"
|
|
dataset = "Tier 1-3"
|
|
source = "https://openai.com/index/introducing-gpt-5-5/"
|
|
date = "2026-04-23"
|
|
|
|
[[benchmarks]]
|
|
name = "FrontierMath"
|
|
score = 27.1
|
|
metric = "accuracy"
|
|
dataset = "Tier 4"
|
|
source = "https://openai.com/index/introducing-gpt-5-5/"
|
|
date = "2026-04-23"
|
|
|
|
[[benchmarks]]
|
|
name = "MMMU Pro"
|
|
score = 81.2
|
|
metric = "accuracy"
|
|
variant = "no tools"
|
|
source = "https://openai.com/index/introducing-gpt-5-5/"
|
|
date = "2026-04-23"
|