Compare · ModelsLive · 2 picked · head to head
GPT-5 vs gpt-oss-120b
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
GPT-5 wins on 20/21 benchmarks
GPT-5 wins 20 of 21 shared benchmarks. Leads in coding · agentic · knowledge.
Category leads
coding·GPT-5agentic·GPT-5knowledge·GPT-5general·GPT-5language·GPT-5math·gpt-oss-120breasoning·GPT-5
Hype vs Reality
Attention vs performance
GPT-5
#100 by perf·#4 by attention
gpt-oss-120b
#127 by perf·no signal
Best value
gpt-oss-120b
50.5x better value than GPT-5
GPT-5
9.4 pts/$
$5.63/M
gpt-oss-120b
475.4 pts/$
$0.10/M
Vendor risk
Who is behind the model
OpenAI
$840.0B·Tier 1
OpenAI
$840.0B·Tier 1
Head to head
21 benchmarks · 2 models
GPT-5gpt-oss-120b
Aider polyglot
GPT-5 leads by +46.2
Aider Polyglot · measures how well AI models can edit code across multiple programming languages using the Aider coding assistant framework.
GPT-5
88.0
gpt-oss-120b
41.8
APEX-Agents
GPT-5 leads by +13.9
APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments.
GPT-5
18.3
gpt-oss-120b
4.4
Chess Puzzles
GPT-5 leads by +17.9
Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities.
GPT-5
33.7
gpt-oss-120b
15.8
Dtbench
GPT-5 leads by +24.0
GPT-5
84.5
gpt-oss-120b
60.5
Fiction.LiveBench
GPT-5 leads by +52.8
Fiction.LiveBench · a continuously updated benchmark using recently published fiction to test reading comprehension and reasoning, preventing data contamination.
GPT-5
97.2
gpt-oss-120b
44.4
GPQA diamond
GPT-5 leads by +13.9
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
GPT-5
81.6
gpt-oss-120b
67.7
HELM · GPQA
GPT-5 leads by +10.7
GPT-5
79.1
gpt-oss-120b
68.4
HELM · IFEval
GPT-5 leads by +3.9
GPT-5
87.5
gpt-oss-120b
83.6
HELM · MMLU-Pro
GPT-5 leads by +6.8
GPT-5
86.3
gpt-oss-120b
79.5
HELM · Omni-MATH
gpt-oss-120b leads by +4.1
GPT-5
64.7
gpt-oss-120b
68.8
HELM · WildBench
GPT-5 leads by +1.2
GPT-5
85.7
gpt-oss-120b
84.5
Lech Mazur Writing
GPT-5 leads by +8.9
Lech Mazur Writing · evaluates creative writing ability, assessing prose quality, narrative coherence, and stylistic sophistication.
GPT-5
86.0
gpt-oss-120b
77.1
Lmca
GPT-5 leads by +21.0
GPT-5
47.0
gpt-oss-120b
26.1
Metr Time Horizons
GPT-5 leads by +13.0
GPT-5
69.6
gpt-oss-120b
56.6
Mystery Game Puzzles
GPT-5 leads by +15.2
GPT-5
15.2
gpt-oss-120b
0.0
OTIS Mock AIME 2024-2025
GPT-5 leads by +2.5
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
GPT-5
91.4
gpt-oss-120b
88.9
SimpleBench
GPT-5 leads by +41.5
SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking.
GPT-5
48.0
gpt-oss-120b
6.5
SimpleQA Verified
GPT-5 leads by +36.2
SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information.
GPT-5
50.1
gpt-oss-120b
13.9
SWE-Bench Verified (Bash Only)
GPT-5 leads by +39.0
SWE-Bench Verified (Bash Only) · a curated subset of SWE-bench where models fix real Python repository bugs using only bash commands, no agent frameworks.
GPT-5
65.0
gpt-oss-120b
26.0
Terminal Bench
GPT-5 leads by +30.9
Terminal-Bench 2.0 · evaluates AI agents on real terminal-based coding tasks · writing scripts, debugging, running tests, and managing projects entirely through command-line interaction. Tests both code quality and terminal fluency. Claude Opus 4.7 scores 69.4%, demonstrating significant agentic terminal competence.
GPT-5
49.6
gpt-oss-120b
18.7
WeirdML
GPT-5 leads by +12.5
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
GPT-5
60.7
gpt-oss-120b
48.2
Full benchmark table
| Benchmark | GPT-5 | gpt-oss-120b |
|---|---|---|
Aider polyglot Aider Polyglot · measures how well AI models can edit code across multiple programming languages using the Aider coding assistant framework. | 88.0 | 41.8 |
APEX-Agents APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments. | 18.3 | 4.4 |
Chess Puzzles Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities. | 33.7 | 15.8 |
Dtbench | 84.5 | 60.5 |
Fiction.LiveBench Fiction.LiveBench · a continuously updated benchmark using recently published fiction to test reading comprehension and reasoning, preventing data contamination. | 97.2 | 44.4 |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 81.6 | 67.7 |
HELM · GPQA | 79.1 | 68.4 |
HELM · IFEval | 87.5 | 83.6 |
HELM · MMLU-Pro | 86.3 | 79.5 |
HELM · Omni-MATH | 64.7 | 68.8 |
HELM · WildBench | 85.7 | 84.5 |
Lech Mazur Writing Lech Mazur Writing · evaluates creative writing ability, assessing prose quality, narrative coherence, and stylistic sophistication. | 86.0 | 77.1 |
Lmca | 47.0 | 26.1 |
Metr Time Horizons | 69.6 | 56.6 |
Mystery Game Puzzles | 15.2 | 0.0 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 91.4 | 88.9 |
SimpleBench SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking. | 48.0 | 6.5 |
SimpleQA Verified SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information. | 50.1 | 13.9 |
SWE-Bench Verified (Bash Only) SWE-Bench Verified (Bash Only) · a curated subset of SWE-bench where models fix real Python repository bugs using only bash commands, no agent frameworks. | 65.0 | 26.0 |
Terminal Bench Terminal-Bench 2.0 · evaluates AI agents on real terminal-based coding tasks · writing scripts, debugging, running tests, and managing projects entirely through command-line interaction. Tests both code quality and terminal fluency. Claude Opus 4.7 scores 69.4%, demonstrating significant agentic terminal competence. | 49.6 | 18.7 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 60.7 | 48.2 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| $1.25 | $10.00 | 400K tokens (~200 books) | $34.38 | |
| $0.04 | $0.17 | 131K tokens (~66 books) | $0.70 |