Codesota · Models · o3OpenAI18 results · 17 benchmarks
Model card

o3.

OpenAIapi5 current SOTA
§ 01 · Benchmarks

Every benchmark o3 has a recorded score for.

#BenchmarkArea · TaskMetricValueRankDateSource
01MMLUReasoning · Commonsense Reasoningaccuracy92.9%#1/412025-04-16source ↗
02RE-BenchAgentic AI · RE-Benchnormalized-score0.4%#1/52025-04-01source ↗
03AIME 2024Reasoning · Mathematical Reasoningaccuracy96.7%#1/8source ↗
04ARC-AGI-1Reasoning · Logical Reasoningaccuracy87.5%#1/5source ↗
05ARC-ChallengeReasoning · Commonsense Reasoningaccuracy98.1%#1/10source ↗
06HCASTAgentic AI · HCASTsuccess-rate49.0%#2/62025-04-01source ↗
07METR Time HorizonAgentic AI · Time Horizontask-horizon-minutes30.0%#2/52025-04-01source ↗
08AIME 2025Reasoning · Mathematical Reasoningaccuracy86.7%#2/5source ↗
09ARC-AGI-2Reasoning · Logical Reasoningaccuracy4.0%#2/3source ↗
10GSM8KReasoning · Mathematical Reasoningaccuracy99.0%#3/32source ↗
11MATHReasoning · Mathematical Reasoningaccuracy97.8%#4/34source ↗
12HumanEvalComputer Code · Code Generationpass@194.8%#5/422025-04-01source ↗
13GPQAReasoning · Multi-step Reasoningaccuracy82.8%#8/33source ↗
14LiveCodeBench ProComputer Code · Code Generationelo1010.00#9/9source ↗
15LiveCodeBenchComputer Code · Code Generationpass@165.3%#11/302024-03-12source ↗
16SWE-Bench VerifiedComputer Code · Code Generationresolve-rate69.1%#21/39source ↗
17HumanEvalComputer Code · Code Generationpass@187.4%#26/42source ↗
18SWE-bench VerifiedAgentic AI · SWE-benchresolve-rate69.1%#44/81source ↗
Rank column shows this model’s position vs all other models scored on the same benchmark + metric (competitors after the slash). #1 in red means current SOTA. Sorted by rank, then newest result.
§ 02 · Strengths by area

Where o3 actually performs.

Reasoning
9
benchmarks
avg rank #2.6 · 4 SOTA
Agentic AI
4
benchmarks
avg rank #12.3 · 1 SOTA
Computer Code
4
benchmarks
avg rank #14.4
§ 03 · Papers

2 papers with results for o3.

  1. 2025-04-01· Agentic AI· 3 results

    METR: Measuring Autonomy in AI Systems (2025 Update)

  2. 2024-03-12· Computer Code· 1 result

    LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code

§ 04 · Related models

Other OpenAI models scored on Codesota.

GPT-4o
Undisclosed params · 35 results · 9 SOTA
o4-mini
13 results · 3 SOTA
o3 (high)
2 results · 1 SOTA
o4-mini (high)
1 result · 1 SOTA
o1
11 results
GPT-5
8 results
o1-preview
Undisclosed params · 8 results
GPT-4.1
7 results
§ 05 · Sources & freshness

Where these numbers come from.

openai-simple-evals
6
results
official-leaderboard
3
results
openai-system-card
2
results
arcprize-leaderboard
2
results
arxiv
1
result
shadow-page-humaneval
1
result
livecodebench-pro-official
1
result
openai-blog
1
result
editorial
1
result
14 of 18 rows marked verified. · first result 2024-03-12, latest 2025-04-16.