o3 vs o4 Mini
o3 leads o4 Mini on the Artificial Analysis Intelligence Index, 20.2 to 16.7, and o4 Mini costs 1.8x less per output token ($4.40 vs $8.00 per million). Updated October 22, 2026; 96 benchmarks have a result for both.
o3 or o4 Mini: which to pick
Neither model has a coding score yet, so the two cannot be ranked on coding.
Pick o3 for work measured by SWE-bench Verified or LiveCodeBench (Artificial Analysis run), where it leads.
- Shared results
- 96
- o3 only
- 6
- o4 Mini only
- 2
- Like-for-like categories
- 8 of 9
Only benchmarks that tested both models count toward a pick; a result only one model has is shown but not compared.
Which one for your work
Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.
Coding
Vibe Coding Index, six public coding benchmarks
No call
Neither has a coding score yet.
One side not measured
Intelligence
AA Intelligence Index
o3
o3 leads, 20.2 against 16.7.
Same board, both measured
Reasoning
GPQA Diamond
o3
o3 leads, 84.1% against 74.5%.
Same board, both measured
Math
AIME
o3
o3 leads, 85.3% against 83.7%.
Same board, both measured
Knowledge
MMLU Pro
o3
o3 leads, 85.6% against 80.6%.
Same board, both measured
Web design
Design Arena: websites
o3
o3 leads, 1036 against 985.
Same board, both measured
Long documents
Listed context window
No call
Both take 200K tokens.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
o4 Mini
o3 $0.0060, o4 Mini $0.0033 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only o3 has the number, $0.571.
Directional only
At a glance
- Coding index
- o3 not benchmarked, o4 Mini not benchmarked.
- Widest category
- Agentic work, 12.2 points apart on average over 3 shared benchmarks.
- Widest single gap
- SimpleQA Verified: o3 49.4% against o4 Mini 19.6%, 29.8 points apart on a 0 to 100 scale.
- Output price
- o3 $8.00 against o4 Mini $4.40 per million output tokens, 1.8x apart.
- Context window
- Both take 200K tokens.
Benchmark by benchmark
First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.
| Category | o3 | Bars | o4 Mini | Ties |
|---|---|---|---|---|
| Coding | 7 | 4 | ||
| Intelligence | 2 | 0 | ||
| Agentic work | 3 | 0 | ||
| Reasoning | 14 | 2 | 1 tie | |
| Math | 5 | 5 | ||
| Knowledge | 14 | 1 | ||
| Web design | 5 | 0 | ||
| General | 33 | 0 | ||
| All shared boards | 83 | 12 | 1 tie |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index74
| Benchmark | o3 | o4 Mini | Leader |
|---|---|---|---|
| Aider polyglot | 81.3% | 72.0% | o3 |
| SWE-bench Verified | 58.4% | 64.6% | o4 Mini |
| SWE-bench Multimodal | 36.0% | 33.9% | o3 |
| LiveCodeBench (Vals AI) | 83.9% | 82.2% | o3 |
| LiveCodeBench (Artificial Analysis run) | 80.8% | 85.9% | o4 Mini |
| Terminal-Bench Hard (Artificial Analysis run) | 37.1% | 15.2% | o3 |
| GSO-Bench | 8.8% | 3.6% | o3 |
| WeirdML | 52.4% | 52.6% | o4 Mini |
| ALE-Bench | 933.5 | Not measured | Only o3 measured |
| LMArena Text: coding | 1407 | 1368 | o3 |
| LMArena Text: software and IT | 1415 | 1367 | o3 |
| SciCode | 9.2% | 10.8% | o4 Mini |
| Neither model has a result on the other 27 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index20
| Benchmark | o3 | o4 Mini | Leader |
|---|---|---|---|
| AA Intelligence Index | 20.2 | 16.7 | o3 |
| Epoch Capabilities Index | 146.7 | 145.5 | o3 |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index30
| Benchmark | o3 | o4 Mini | Leader |
|---|---|---|---|
| τ²-Bench Telecom | 80.7% | 55.6% | o3 |
| METR time horizons | 65.4% | 63.9% | o3 |
| GDPval | 30.8% | Not measured | Only o3 measured |
| DeepResearch Bench | 45.2% | Not measured | Only o3 measured |
| OSWorld | 23.0% | Not measured | Only o3 measured |
| Berkeley Function Calling Leaderboard (BFCL) | 63.0% | 53.2% | o3 |
| BrowseComp-Plus | 65.9% | Not measured | Only o3 measured |
| Neither model has a result on the other 34 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index142
| Benchmark | o3 | o4 Mini | Leader |
|---|---|---|---|
| GPQA Diamond | 84.1% | 74.5% | o3 |
| MMMU Pro (Vals AI) | 80.4% | 79.7% | o3 |
| GPQA Diamond (Artificial Analysis run) | 82.7% | 78.4% | o3 |
| Humanity's Last Exam (Artificial Analysis run) | 20.1% | 16.5% | o3 |
| GPQA Diamond (Epoch AI run) | 81.8% | 79.6% | o3 |
| ARC-AGI | 60.8% | 58.7% | o3 |
| ARC-AGI-2 | 6.5% | 6.1% | o3 |
| SimpleBench | 53.1% | 38.7% | o3 |
| Chess puzzles | 37.0% | 22.0% | o3 |
| Mystery game puzzles | 29.0% | 5.0% | o3 |
| VPCT | 52.0% | 57.5% | o4 Mini |
| LMArena Text: hard prompts | 1402 | 1351 | o3 |
| LMArena Text: expert | 1402 | 1343 | o3 |
| LMArena Text: hard prompts in English | 1411 | 1366 | o3 |
| LMArena Vision: diagrams | 1220 | 1197 | o3 |
| LMArena Vision: homework | 1234 | 1234 | o4 Mini |
| Kagi LLM Benchmark | 67.6% | 67.6% | Tie |
| Neither model has a result on the other 5 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index55
| Benchmark | o3 | o4 Mini | Leader |
|---|---|---|---|
| AIME | 85.3% | 83.7% | o3 |
| MATH 500 (Vals AI) | 94.6% | 94.2% | o3 |
| MGSM (Vals AI) | 91.8% | 93.4% | o4 Mini |
| AA Math Index | 88.3 | 90.7 | o4 Mini |
| AIME 2025 (Artificial Analysis run) | 88.3% | 90.7% | o4 Mini |
| OTIS Mock AIME | 84.4% | 81.7% | o3 |
| FrontierMath, tiers 1 to 3 | 33.3% | 36.1% | o4 Mini |
| FrontierMath, tier 4 | Not measured | 4.9% | Only o4 Mini measured |
| MATH level 5 | 97.8% | 97.8% | o4 Mini |
| LMArena Text: math | 1426 | 1389 | o3 |
| LMArena Text: mathematical work | 1423 | 1379 | o3 |
| Neither model has a result on the other 2 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index141
| Benchmark | o3 | o4 Mini | Leader |
|---|---|---|---|
| MMLU Pro | 85.6% | 80.6% | o3 |
| LegalBench (Vals AI) | 83.8% | 79.2% | o3 |
| CorpFin (Vals AI) | 59.7% | 59.0% | o3 |
| TaxEval (Vals AI) | 74.6% | 74.8% | o4 Mini |
| MortgageTax (Vals AI) | 65.7% | 64.8% | o3 |
| MedQA (Vals AI) | 96.1% | 96.0% | o3 |
| MedCode (Vals AI) | 47.3% | 33.8% | o3 |
| MMLU Pro (Artificial Analysis run) | 85.3% | 83.2% | o3 |
| SimpleQA Verified | 49.4% | 19.6% | o3 |
| GeoBench | 74.0% | 64.0% | o3 |
| LMArena Text: science | 1429 | 1354 | o3 |
| LMArena Text: medicine and healthcare | 1448 | 1361 | o3 |
| LMArena Text: legal and government | 1428 | 1350 | o3 |
| LMArena Text: business and finance | 1393 | 1325 | o3 |
| FACTS Benchmark Suite | 44.4% | 28.1% | o3 |
| Neither model has a result on the other 3 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index50
| Benchmark | o3 | o4 Mini | Leader |
|---|---|---|---|
| Design Arena: websites | 1036 | 985 | o3 |
| Design Arena: UI components | 1017 | 983 | o3 |
| Design Arena: game dev | 1034 | 1003 | o3 |
| Design Arena: data viz | 1200 | 991 | o3 |
| Design Arena: 3D | Not measured | 854 | Only o4 Mini measured |
| Design Arena: all code categories | 1020 | 976 | o3 |
| Neither model has a result on the other 3 boards here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index330
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | o3 | o4 Mini | Leader |
|---|---|---|---|
| Neither model has a result on 5 boards here | |||
What each workload costs
Three fixed token mixes at each model's listed price, so per-token rates become a bill.
Chat turn
1K input + 500 output tokens
o4 Mini costs less
Repository review
50K input + 3K output tokens
o4 Mini costs less
Agent loop
200K input + 10K output tokens
o4 Mini costs less
Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.
Specification differences
The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.
| Specification | OpenAI o3 | OpenAI o4 Mini |
|---|---|---|
| Maker | OpenAI | OpenAI |
| Weights | No public weights listed | No public weights listed |
| Released | April 16, 2025 | April 16, 2025 |
| Context window | 200K | 200K |
| Input $/Mtok | $2.00 | $1.10 |
| Output $/Mtok | $8.00 | $4.40 |
| Coding index | Not benchmarked | Not benchmarked |
| Independent measurements behind it | 0 | 0 |
| Cost per solved task | $0.571 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is o3 better than o4 Mini for coding?
Neither model has a coding score yet, so the two cannot be ranked on coding. Pick o3 for work measured by SWE-bench Verified or LiveCodeBench (Artificial Analysis run), where it leads.
Which is cheaper, o3 or o4 Mini?
o3 costs $2.00 per million input tokens and $8.00 per million output tokens; o4 Mini costs $1.10 and $4.40. Cost per solved task is known for o3 only.
Which has the bigger context window, o3 or o4 Mini?
Neither: both take 200K tokens.
How many benchmarks is the o3 vs o4 Mini comparison based on?
96 benchmarks have a result for both models as of October 22, 2026. o3's coding index rests on 0 independent benchmarks and o4 Mini's on 0. Where only one model has a result, the page says so instead of naming a winner.