Gemini 4 Carbon vs GPT-5 Mini
Gemini 4 Carbon leads GPT-5 Mini on coding, 86.6 to 34.2 on the Vibe Coding Index, and GPT-5 Mini costs 5.0x less per output token ($2.00 vs $10.00 per million). Updated November 14, 2026; 42 benchmarks have a result for both.
Gemini 4 Carbon or GPT-5 Mini: which to pick
Pick Gemini 4 Carbon for coding: it scores 86.6 on the Vibe Coding Index against 34.2 for GPT-5 Mini.
Pick GPT-5 Mini for output tokens 5.0x cheaper ($2.00 vs $10.00 per million).
- Shared results
- 42
- Gemini 4 Carbon only
- 31
- GPT-5 Mini only
- 61
- 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
Gemini 4 Carbon
Gemini 4 Carbon scores 86.6 against 34.2.
4 and 2 benchmarks behind the scores
Intelligence
AA Intelligence Index
No call
Only Gemini 4 Carbon has a AA Intelligence Index result, 60.0.
Directional only
Agentic work
AA Agentic Index
No call
Only GPT-5 Mini has a AA Agentic Index result, 6.8.
Directional only
Reasoning
GPQA Diamond
No call
Only GPT-5 Mini has a GPQA Diamond result, 80.3%.
Directional only
Math
AIME
No call
Only GPT-5 Mini has a AIME result, 91.5%.
Directional only
Knowledge
MMLU Pro
No call
Only GPT-5 Mini has a MMLU Pro result, 82.2%.
Directional only
Web design
Design Arena: websites
Gemini 4 Carbon
Gemini 4 Carbon leads, 1319 against 1117.
Same board, both measured
General
Vals Index
No call
Only Gemini 4 Carbon has a Vals Index result, 70.5%.
Directional only
Long documents
Listed context window
Gemini 4 Carbon
Gemini 4 Carbon takes 1.05M tokens against 400K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
GPT-5 Mini
Gemini 4 Carbon $0.0070, GPT-5 Mini $0.0013 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only GPT-5 Mini has the number, $0.08.
Directional only
At a glance
- Coding index
- Gemini 4 Carbon 86.6 from 4 benchmarks, GPT-5 Mini 34.2 from 2 benchmarks.
- Widest category
- Agentic work, 100.0 points apart on average over 1 shared benchmark.
- Widest single gap
- Vending-Bench 2: Gemini 4 Carbon $16,484 against GPT-5 Mini -$31, 100.0 points apart on a 0 to 100 scale.
- Output price
- Gemini 4 Carbon $10.00 against GPT-5 Mini $2.00 per million output tokens, 5.0x apart.
- Context window
- Gemini 4 Carbon takes 1.05M tokens, GPT-5 Mini 400K.
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 | Gemini 4 Carbon | Bars | GPT-5 Mini | Ties |
|---|---|---|---|---|
| Coding | 4 | 0 | ||
| Intelligence | 1 | 0 | ||
| Agentic work | 1 | 0 | ||
| Reasoning | 8 | 0 | ||
| Math | 5 | 0 | ||
| Knowledge | 5 | 0 | ||
| Web design | 5 | 0 | ||
| General | 13 | 0 | ||
| All shared boards | 42 | 0 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index40
| Benchmark | Gemini 4 Carbon | GPT-5 Mini | Leader |
|---|---|---|---|
| SWE-bench Verified | Not measured | 59.8% | Only GPT-5 Mini measured |
| SWE-bench Multilingual | Not measured | 39.7% | Only GPT-5 Mini measured |
| LiveBench coding | 91.0% | Not measured | Only Gemini 4 Carbon measured |
| LMArena WebDev | 1768 | Not measured | Only Gemini 4 Carbon measured |
| AA Coding Index | Not measured | 15.6 | Only GPT-5 Mini measured |
| LiveCodeBench (Vals AI) | Not measured | 86.6% | Only GPT-5 Mini measured |
| SWE-bench Verified (Vals AI) | Not measured | 60.8% | Only GPT-5 Mini measured |
| Terminal-Bench 4 (Vals AI) | 63.1% | Not measured | Only Gemini 4 Carbon measured |
| Vibe Code Bench (Vals AI) | 89.1% | 14.2% | Gemini 4 Carbon |
| LiveCodeBench (Artificial Analysis run) | Not measured | 83.8% | Only GPT-5 Mini measured |
| SciCode | Not measured | 39.0% | Only GPT-5 Mini measured |
| Terminal-Bench Hard (Artificial Analysis run) | Not measured | 33.3% | Only GPT-5 Mini measured |
| Terminal-Bench 4.0 (Artificial Analysis run) | 65.2% | 0.0% | Gemini 4 Carbon |
| FrontierCode 2.0 | 40.6% | Not measured | Only Gemini 4 Carbon measured |
| Terminal-Bench | Not measured | 34.8% | Only GPT-5 Mini measured |
| FrontierSWE | 61.2% | Not measured | Only Gemini 4 Carbon measured |
| WeirdML | Not measured | 52.7% | Only GPT-5 Mini measured |
| ALE-Bench | 2374.7 | Not measured | Only Gemini 4 Carbon measured |
| LMArena Text: coding | 1564 | 1406 | Gemini 4 Carbon |
| LMArena Text: software and IT | 1552 | 1394 | Gemini 4 Carbon |
| LMArena WebDev: React | 1792 | Not measured | Only Gemini 4 Carbon measured |
| LMArena WebDev: HTML | 1770 | Not measured | Only Gemini 4 Carbon measured |
| Terminal-Bench 2.0 | Not measured | 34.8% | Only GPT-5 Mini measured |
| CursorBench 4.0 | 60.4% | Not measured | Only Gemini 4 Carbon measured |
| ProgramBench | Not measured | 0.0% | Only GPT-5 Mini measured |
| Neither model has a result on the other 27 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | Gemini 4 Carbon | GPT-5 Mini | Leader |
|---|---|---|---|
| AA Intelligence Index | 60.0 | Not measured | Only Gemini 4 Carbon measured |
| Epoch Capabilities Index | 169.8 | 146.1 | Gemini 4 Carbon |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | Gemini 4 Carbon | GPT-5 Mini | Leader |
|---|---|---|---|
| AA Agentic Index | Not measured | 6.8 | Only GPT-5 Mini measured |
| Design Arena: agents building mobile apps | 1284 | Not measured | Only Gemini 4 Carbon measured |
| Design Arena: spreadsheets | 1224 | Not measured | Only Gemini 4 Carbon measured |
| LiveBench research | 92.5% | Not measured | Only Gemini 4 Carbon measured |
| LiveBench browser | 80.0% | Not measured | Only Gemini 4 Carbon measured |
| LiveBench long agentic | 72.7% | Not measured | Only Gemini 4 Carbon measured |
| Tax Agent Bench (Vals AI) | 51.8% | Not measured | Only Gemini 4 Carbon measured |
| Legal Research Bench (Vals AI) | 55.8% | Not measured | Only Gemini 4 Carbon measured |
| Harvey's Legal Agent Benchmark (Vals AI) | 17.5% | Not measured | Only Gemini 4 Carbon measured |
| τ²-Bench Telecom | Not measured | 71.1% | Only GPT-5 Mini measured |
| APEX-Agents | 84.2% | Not measured | Only Gemini 4 Carbon measured |
| Vending-Bench 2 | $16,484 | -$31 | Gemini 4 Carbon |
| Berkeley Function Calling Leaderboard (BFCL) | Not measured | 55.5% | Only GPT-5 Mini measured |
| Neither model has a result on the other 31 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index80
| Benchmark | Gemini 4 Carbon | GPT-5 Mini | Leader |
|---|---|---|---|
| GPQA Diamond | Not measured | 80.3% | Only GPT-5 Mini measured |
| LiveBench reasoning | 90.6% | Not measured | Only Gemini 4 Carbon measured |
| MMMU Pro (Vals AI) | Not measured | 78.9% | Only GPT-5 Mini measured |
| GPQA Diamond (Artificial Analysis run) | Not measured | 82.8% | Only GPT-5 Mini measured |
| Humanity's Last Exam (Artificial Analysis run) | Not measured | 21.5% | Only GPT-5 Mini measured |
| HLE-Diamond (Artificial Analysis run) | 69.1% | Not measured | Only Gemini 4 Carbon measured |
| CritPt v2, hard subset (Artificial Analysis run) | 69.0% | Not measured | Only Gemini 4 Carbon measured |
| GPQA Diamond (Epoch AI run) | 96.5% | 75.0% | Gemini 4 Carbon |
| ARC-AGI | 98.5% | 54.3% | Gemini 4 Carbon |
| ARC-AGI-2 | 94.2% | 4.4% | Gemini 4 Carbon |
| SimpleBench | 92.6% | Not measured | Only Gemini 4 Carbon measured |
| Chess puzzles | 78.0% | 24.0% | Gemini 4 Carbon |
| Mystery game puzzles | 73.0% | 10.0% | Gemini 4 Carbon |
| EBR-bench | 87.6% | Not measured | Only Gemini 4 Carbon measured |
| VPCT | Not measured | 40.2% | Only GPT-5 Mini measured |
| LMArena Text: hard prompts | 1551 | 1380 | Gemini 4 Carbon |
| LMArena Text: expert | 1582 | 1379 | Gemini 4 Carbon |
| LMArena Text: hard prompts in English | 1545 | 1383 | Gemini 4 Carbon |
| LMArena Vision: diagrams | Not measured | 1202 | Only GPT-5 Mini measured |
| LMArena Vision: homework | Not measured | 1212 | Only GPT-5 Mini measured |
| Kagi LLM Benchmark | Not measured | 70.3% | Only GPT-5 Mini measured |
| Neither model has a result on the other 5 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index50
| Benchmark | Gemini 4 Carbon | GPT-5 Mini | Leader |
|---|---|---|---|
| AIME | Not measured | 91.5% | Only GPT-5 Mini measured |
| MATH 500 (Vals AI) | Not measured | 94.8% | Only GPT-5 Mini measured |
| MGSM (Vals AI) | Not measured | 92.6% | Only GPT-5 Mini measured |
| AIME 2025 (Artificial Analysis run) | Not measured | 90.7% | Only GPT-5 Mini measured |
| OTIS Mock AIME | 100.0% | 86.7% | Gemini 4 Carbon |
| FrontierMath, tiers 1 to 3 | 90.5% | 46.7% | Gemini 4 Carbon |
| FrontierMath, tier 4 | 95.1% | 12.2% | Gemini 4 Carbon |
| ProofBench | Not measured | 9.0% | Only GPT-5 Mini measured |
| MATH level 5 | Not measured | 97.8% | Only GPT-5 Mini measured |
| LMArena Text: math | Not measured | 1378 | Only GPT-5 Mini measured |
| LMArena Text: mathematical work | 1538 | 1381 | Gemini 4 Carbon |
| MathArena | 85.4% | 15.0% | Gemini 4 Carbon |
| Neither model has a result on the other 2 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index50
| Benchmark | Gemini 4 Carbon | GPT-5 Mini | Leader |
|---|---|---|---|
| MMLU Pro | Not measured | 82.2% | Only GPT-5 Mini measured |
| LegalBench (Vals AI) | Not measured | 81.8% | Only GPT-5 Mini measured |
| CaseLaw (Vals AI) | Not measured | 68.5% | Only GPT-5 Mini measured |
| CorpFin (Vals AI) | Not measured | 60.2% | Only GPT-5 Mini measured |
| TaxEval (Vals AI) | Not measured | 75.2% | Only GPT-5 Mini measured |
| MortgageTax (Vals AI) | Not measured | 66.9% | Only GPT-5 Mini measured |
| MedQA (Vals AI) | Not measured | 96.1% | Only GPT-5 Mini measured |
| MedCode (Vals AI) | Not measured | 43.0% | Only GPT-5 Mini measured |
| AA-Omniscience | 53.5 | Not measured | Only Gemini 4 Carbon measured |
| SimpleQA Verified | 77.7% | 21.6% | Gemini 4 Carbon |
| LMArena Text: science | 1535 | 1370 | Gemini 4 Carbon |
| LMArena Text: medicine and healthcare | 1525 | 1380 | Gemini 4 Carbon |
| LMArena Text: legal and government | 1542 | 1375 | Gemini 4 Carbon |
| LMArena Text: business and finance | 1520 | 1358 | Gemini 4 Carbon |
| FACTS Benchmark Suite | 78.4% | Not measured | Only Gemini 4 Carbon measured |
| PRBench Legal | 53.3% | Not measured | Only Gemini 4 Carbon measured |
| Neither model has a result on the other 5 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index50
| Benchmark | Gemini 4 Carbon | GPT-5 Mini | Leader |
|---|---|---|---|
| Design Arena: websites | 1319 | 1117 | Gemini 4 Carbon |
| Design Arena: UI components | 1357 | 1096 | Gemini 4 Carbon |
| Design Arena: game dev | 1326 | 1119 | Gemini 4 Carbon |
| Design Arena: data viz | Not measured | 1127 | Only GPT-5 Mini measured |
| Design Arena: 3D | 1434 | 1017 | Gemini 4 Carbon |
| Design Arena: all code categories | 1348 | 1104 | Gemini 4 Carbon |
| Design Arena: SVG | Not measured | 1080 | Only GPT-5 Mini measured |
| Design Arena: ASCII art | Not measured | 1119 | Only GPT-5 Mini measured |
| LiveBench design | 80.2% | Not measured | Only Gemini 4 Carbon measured |
GeneralLike-for-likePublic boards, shown as published, kept out of the index130
| Benchmark | Gemini 4 Carbon | GPT-5 Mini | Leader |
|---|---|---|---|
| Vals Index | 70.5% | Not measured | Only Gemini 4 Carbon measured |
| LiveBench data analysis | 86.5% | Not measured | Only Gemini 4 Carbon measured |
| MedScribe (Vals AI) | Not measured | 80.6% | Only GPT-5 Mini measured |
| IFBench | Not measured | 75.4% | Only GPT-5 Mini measured |
| Long context reasoning (AA-LCR) | 79.3% | 72.3% | Gemini 4 Carbon |
| GDPval-AA | 1733 | 773 | Gemini 4 Carbon |
| DTBench | Not measured | 80.5% | Only GPT-5 Mini measured |
| LMCA | Not measured | 34.2% | Only GPT-5 Mini measured |
| Fiction.LiveBench | Not measured | 69.4% | Only GPT-5 Mini measured |
| Furniture assembly | 77.5% | Not measured | Only Gemini 4 Carbon measured |
| LMArena Text | 1538 | 1373 | Gemini 4 Carbon |
| LMArena Text: instruction following | 1539 | 1357 | Gemini 4 Carbon |
| LMArena Text: creative writing | 1533 | 1325 | Gemini 4 Carbon |
| LMArena Vision | Not measured | 1202 | Only GPT-5 Mini measured |
| LMArena Text: English | 1530 | 1379 | Gemini 4 Carbon |
| LMArena Text: non-English | 1534 | 1363 | Gemini 4 Carbon |
| LMArena Text: Chinese | 1613 | 1386 | Gemini 4 Carbon |
| LMArena Text: German | Not measured | 1365 | Only GPT-5 Mini measured |
| LMArena Text: Japanese | Not measured | 1341 | Only GPT-5 Mini measured |
| LMArena Text: Korean | Not measured | 1304 | Only GPT-5 Mini measured |
| LMArena Text: French | Not measured | 1386 | Only GPT-5 Mini measured |
| LMArena Text: Spanish | Not measured | 1355 | Only GPT-5 Mini measured |
| LMArena Text: Russian | 1560 | 1362 | Gemini 4 Carbon |
| LMArena Text: Polish | Not measured | 1378 | Only GPT-5 Mini measured |
| LMArena Text: multi-turn | 1569 | 1363 | Gemini 4 Carbon |
| LMArena Text: longer queries | 1549 | 1355 | Gemini 4 Carbon |
| LMArena Text: writing and language | 1542 | 1344 | Gemini 4 Carbon |
| LMArena Text: entertainment and media | 1527 | 1340 | Gemini 4 Carbon |
| LMArena Vision: English | Not measured | 1216 | Only GPT-5 Mini measured |
| LMArena Vision: Chinese | Not measured | 1214 | Only GPT-5 Mini measured |
| LMArena Vision: reading text | Not measured | 1199 | Only GPT-5 Mini measured |
| LMArena Vision: humor | Not measured | 1205 | Only GPT-5 Mini measured |
| LMArena Vision: captioning | Not measured | 1216 | Only GPT-5 Mini measured |
| LMArena Vision: recognizing things | Not measured | 1226 | Only GPT-5 Mini measured |
| LMArena Vision: creative writing | Not measured | 1182 | Only GPT-5 Mini measured |
| MultiChallenge | Not measured | 59.0% | Only GPT-5 Mini measured |
| Neither model has a result on the other 13 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Gemini 4 Carbon | GPT-5 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
GPT-5 Mini costs less
Repository review
50K input + 3K output tokens
GPT-5 Mini costs less
Agent loop
200K input + 10K output tokens
GPT-5 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 | Google Gemini 4 Carbon | OpenAI GPT-5 Mini |
|---|---|---|
| Maker | OpenAI | |
| Weights | No public weights listed | No public weights listed |
| Released | November 10, 2026 | August 7, 2025 |
| Context window | 1.05M | 400K |
| Input $/Mtok | $2.00 | $0.25 |
| Output $/Mtok | $10.00 | $2.00 |
| Coding index | 86.6 | 34.2 |
| Independent measurements behind it | 4 | 2 |
| Cost per solved task | Not computable | $0.08 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Gemini 4 Carbon better than GPT-5 Mini for coding?
Pick Gemini 4 Carbon for coding: it scores 86.6 on the Vibe Coding Index against 34.2 for GPT-5 Mini. Pick GPT-5 Mini for output tokens 5.0x cheaper ($2.00 vs $10.00 per million).
Which is cheaper, Gemini 4 Carbon or GPT-5 Mini?
Gemini 4 Carbon costs $2.00 per million input tokens and $10.00 per million output tokens; GPT-5 Mini costs $0.25 and $2.00. Cost per solved task is known for GPT-5 Mini only.
Which has the bigger context window, Gemini 4 Carbon or GPT-5 Mini?
Gemini 4 Carbon, with 1.05M tokens against 400K.
How many benchmarks is the Gemini 4 Carbon vs GPT-5 Mini comparison based on?
42 benchmarks have a result for both models as of November 14, 2026. Gemini 4 Carbon's coding index rests on 4 independent benchmarks and GPT-5 Mini's on 2. Where only one model has a result, the page says so instead of naming a winner.