Gemini 4 Carbon vs GPT-5.4
Full profiles: Gemini 4 Carbon (Google) and GPT-5.4 (OpenAI): every benchmark, price and rank.
Updated · Data from 10 sources
As of 29 Oct 2026, Gemini 4 Carbon and GPT-5.4 are neck and neck on quality (65.1 vs 63.3 on the BenchLeader Index, inside the ±6.6 margin), and GPT-5.4 is 1.4× cheaper at $5.63 vs $8.00 per 1M tokens blended. Choose Gemini 4 Carbon for knowledge (76 vs 59), reasoning (79 vs 66), and coding (72 vs 65); choose GPT-5.4 for long context (66 vs 65).
Key facts (as of 29 Oct 2026)
- Gemini 4 Carbon scores 65.1 on the BenchLeader Index and GPT-5.4 scores 63.3 (BenchLeader, 29 Oct 2026).
- Gemini 4 Carbon costs $8.00 per 1M tokens blended (list price, 29 Oct 2026).
- GPT-5.4 costs $5.63 per 1M tokens blended (list price, 29 Oct 2026).
- Of the 6 benchmarks both models have, Gemini 4 Carbon leads 5 (BenchLeader, 29 Oct 2026).
Gemini 4 Carbon benchmarks 
GPT-5.4 benchmarks Which to pick for…
The pick's numbers come first. Category scores on the BenchLeader scale (50 is the average model, 15 points one standard deviation). High confidence means a lead of half a standard deviation or more; under one point is called either.
| Use case | Pick | Why | Confidence |
|---|---|---|---|
| Coding | Gemini 4 Carbon | 72 vs 65 category score; SciCode 64.2% vs 56.6% | Medium |
| Reasoning | Gemini 4 Carbon | 79 vs 66 category score; Humanity's Last Exam (AA) 60.4% vs 43.7% | High |
| Knowledge | Gemini 4 Carbon | 76 vs 59 category score | High |
| Long documents | GPT-5.4 | 66 vs 65 category score; AA-LCR v1.1 82.0% vs 79.3% | Medium |
| Lowest cost | GPT-5.4 | $5.63 vs $8.00 per 1M blended; agentic coding session $0.210 vs $0.320 | – |
| Fastest first answer | Gemini 4 Carbon | 103.1 s (Gemini 4 Carbon (high)) vs 148.2 s (GPT-5.4 (xhigh)) to the first answer token | – |
| High volume on a budget | Neither flagship | A smaller model is usually the better fit: Cheapest good LLMs, Fastest LLMs | – |
Key differences
- Gemini 4 Carbon (high) and GPT-5.4 (xhigh) are neck and neck on quality (65.1 vs 63.3, within the ±6.6 margin).
- Gemini 4 Carbon (high) is stronger in coding, knowledge, reasoning.
- GPT-5.4 (xhigh) is stronger in long context, agents & tools, instruction following, maths, multimodal.
- GPT-5.4 (xhigh) is 1.4× cheaper ($5.63 vs $8.00 per 1M blended).
At a glance
| Spec | Gemini 4 Carbon | GPT-5.4 |
|---|---|---|
| Developer | OpenAI | |
| Released | 28 Oct 2026 | 5 Mar 2026 |
| Context window (tokens) | 1,000,000 | 1,050,000 |
| Compared at | Gemini 4 Carbon (high) | GPT-5.4 (xhigh) |
| Open weights | No | No |
| Price in / out / cached input per 1M | $4.00 / $20.00 / – | $2.50 / $15.00 / $0.250 |
Category by category
Category scores on the BenchLeader scale: 50 is the average evaluated model, 15 points is one standard deviation. A gap under one point is called level.
| Category | Gemini 4 Carbon (high) | GPT-5.4 (xhigh) | Leader |
|---|---|---|---|
| Reasoning | 78.9 | 66.1 | Gemini 4 Carbon |
| Coding | 71.7 | 64.5 | Gemini 4 Carbon |
| Agents & tools | – | 63.6 | not comparable |
| Maths | – | 65.1 | not comparable |
| Knowledge | 75.8 | 58.6 | Gemini 4 Carbon |
| Instruction following | – | 72.3 | not comparable |
| Multimodal | – | 61.1 | not comparable |
| Long context | 64.5 | 65.9 | GPT-5.4 |
Where sources disagree
Across the 6 benchmarks both have, Gemini 4 Carbon leads 5, GPT-5.4 leads 1.
| Overall index | Gemini 4 Carbon (high) | GPT-5.4 (xhigh) | Leader |
|---|---|---|---|
| BenchLeader Index | 65.1 | 63.3 | neck and neck |
Each publisher runs its own tasks, harnesses and effort settings, so their overall verdicts can differ; BenchLeader reconciles them on one scale. † measured at another effort setting of the same model (the publisher did not test the one compared here).
Price, speed and latency
List prices blended at three input tokens per output token. Per-task prices use the token counts on the methodology page; suite cost is what Artificial Analysis paid to run its whole Intelligence Index.
| Measure | Gemini 4 Carbon (high) | GPT-5.4 (xhigh) |
|---|---|---|
| Blended price per 1M tokens | $8.00 | $5.63 |
| Input / output per 1M | $4.00 / $20.00 | $2.50 / $15.00 |
| Cached input per 1M | – | $0.250 |
| Price per task: chat reply | $0.0076 | $0.0055 |
| Price per task: summarize a 30-page report | $0.060 | $0.039 |
| Price per task: code edit | $0.054 | $0.037 |
| Price per task: agentic coding session | $0.320 | $0.210 |
| Price per task: structured extraction | $0.012 | $0.0080 |
| Cost to run a full benchmark suite | $7.76 | – |
| Output speed | 100 tok/s | 102 tok/s |
| Time to first answer | 103.1 s | 148.2 s |
| Time to a complete answer | 108.1 s | 153.0 s |
Each model is compared at its best-ranked reasoning-effort setting. Gemini 4 Carbon (high) scores 65.1 and takes 103 s to its first answer token. GPT-5.4 (xhigh) scores 63.3 and takes 148 s to its first answer token; GPT-5.4 (low) scores 58.9 in 2 s, GPT-5.4 (high) scores 58.9 in 1 s. Higher effort means more thinking before the answer: better scores, longer waits and more tokens billed. How reasoning effort changes scores and cost.
Every benchmark
131 benchmarks, grouped by category; each name links to the board and its leaders. † measured at another effort setting of the same model, shown for reference and not counted as a win.
| Metric | Publisher | Gemini 4 Carbon (high) | GPT-5.4 (xhigh) |
|---|---|---|---|
| BenchLeader Index | BenchLeader | 65.1 | 63.3 |
| Coding score | BenchLeader | 71.7 | 64.5 |
| Knowledge score | BenchLeader | 75.8 | 58.6 |
| Long context score | BenchLeader | 64.5 | 65.9 |
| Reasoning score | BenchLeader | 78.9 | 66.1 |
| Agents & tools score | BenchLeader | – | 63.6 |
| Instruction following score | BenchLeader | – | 72.3 |
| Maths score | BenchLeader | – | 65.1 |
| Multimodal score | BenchLeader | – | 61.1 |
| Blended price $/M | BenchLeader | $8.00 | $5.63 |
| Output speed | BenchLeader | 100 tok/s | 102 tok/s |
| Time to first answer | BenchLeader | 103.1 s | 148.2 s |
| Context window | BenchLeader | 1,000,000 tokens | 1,050,000 tokens |
| Reasoning | |||
| GPQA Diamond | Epoch AI Benchmarking Hub | – | 93.3% |
| Humanity's Last Exam | Scale AI / CAIS | – | 36.2% |
| Arena Hard Prompts | Arena (LMArena) | – | 1487† |
| LiveBench 1.0 Reasoning | LiveBench | – | 88.1% |
| GPQA Diamond (AA) | Artificial Analysis | – | 92.0% |
| Humanity's Last Exam (AA) | Artificial Analysis | 60.4% | 43.7% |
| HLE-Diamond (AA) | Artificial Analysis | 69.1% | – |
| GPQA Diamond (Vals) | Vals AI | – | 91.7% |
| Kagi LLM Benchmark | Kagi LLM Benchmark | – | 63.8%† |
| ARC-AGI-1 | ARC Prize | – | 93.7% |
| ARC-AGI-2 | ARC Prize | – | 74.0% |
| ARC-AGI-3 | ARC Prize | – | 0.2%† |
| CritPt | Artificial Analysis | 36.0% | 23.4% |
| CritPt v2 hard subset (AA) | Artificial Analysis | 63.0% | – |
| Chess Puzzles | Epoch AI Benchmarking Hub | – | 48.0% |
| Mystery Game Puzzles | Epoch AI Benchmarking Hub | – | 37.0% |
| CL-bench | Epoch AI Benchmarking Hub | – | 27.9% |
| CL-bench Life | Epoch AI Benchmarking Hub | – | 21.7% |
| LMCA | Epoch AI Benchmarking Hub | – | 52.0% |
| DTBench | Epoch AI Benchmarking Hub | – | 94.4% |
| ForecastBench | ForecastBench | – | 63.1%† |
| Coding | |||
| SWE-bench Verified (Epoch) | Epoch AI Benchmarking Hub | – | 76.9%† |
| SciCode | SciCode | 64.2% | 56.6% |
| WeirdML | WeirdML | – | 77.7% |
| GSO-Bench | GSO-Bench | – | 31.4% |
Show all 131 benchmarks
| Metric | Publisher | Gemini 4 Carbon (high) | GPT-5.4 (xhigh) |
|---|---|---|---|
| Coding | |||
| Arena Coding | Arena (LMArena) | – | 1511† |
| Arena WebDev | Arena (LMArena) | – | 1401† |
| LiveBench 1.0 Coding | LiveBench | – | 77.5% |
| SciCode (AA) | Artificial Analysis | 64.2% | – |
| LiveCodeBench | Vals AI | – | 84.1% |
| SWE-bench (Vals) | Vals AI | – | 78.2% |
| SWE-Bench Pro | Scale AI SEAL | – | 59.1% |
| Code Migration | Vals AI | – | 35.0% |
| ProgramBench | Vals AI | – | 0.0% |
| Vibe Code Bench v1.1 | Vals AI | 89.1% | 67.4% |
| SWE-Bench Pro (private) | Scale AI SEAL | – | 59.1% |
| SWE Atlas: Codebase QnA | Scale AI SEAL | – | 40.8% |
| SWE Atlas: Refactoring | Scale AI SEAL | – | 44.3% |
| SWE Atlas: Test Writing | Scale AI SEAL | – | 44.4% |
| MirrorCode | Epoch AI Benchmarking Hub | – | 15.6%† |
| DeepSWE v1.1 | Epoch AI Benchmarking Hub | – | 51.8% |
| ALE-Bench | Epoch AI Benchmarking Hub | – | 1607† |
| AlgoTune | AlgoTune | – | 1.9† |
| Terminal-Bench 4.0 (AA) | Artificial Analysis | 65.2% | – |
| Terminal-Bench 2.1 (AA) | Artificial Analysis | – | 78.3% |
| FrontierCode 1.1 (AA) | Artificial Analysis | 48.5% | – |
| FrontierCode 2.0 (AA) | Artificial Analysis | 40.6% | – |
| Agents & tools | |||
| Terminal-Bench | Terminal-Bench | – | 81.8%† |
| APEX-Agents | Mercor | – | 52.4%† |
| LiveBench Agentic Coding | LiveBench | – | 53.8% |
| Terminal-Bench Hard | Artificial Analysis | – | 57.6% |
| τ²-Bench Telecom (AA) | Artificial Analysis | – | 87.1% |
| MCP Atlas | Scale AI SEAL | – | 70.6% |
| HiL-Bench | Scale AI SEAL | – | 9.7%† |
| τ³-bench Banking | τ²-bench | – | 39.4% |
| τ³-Banking (AA) | Artificial Analysis | – | 39.6% |
| APEX-Agents (AA) | Artificial Analysis | – | 33.3% |
| Finance Agent v2.1 | Vals AI | 67.2% | – |
| Harvey's Legal Agent Benchmark v1.1 | Vals AI | 17.5% | – |
| Harvey's Legal Agent Benchmark v1 | Vals AI | – | 0.0% |
| Legal Research Bench | Vals AI | 55.8% | – |
| SkillsBench | Vals AI | – | 51.7% |
| Tax Agent Bench | Vals AI | 48.9% | – |
| Terminal-Bench 2.0 (Vals) | Vals AI | – | 58.4%† |
| Terminal-Bench 4.0 (Vals) | Vals AI | 63.1% | – |
| DeepResearch Bench | DeepResearch Bench | – | 35.1%† |
| PostTrainBench | Epoch AI Benchmarking Hub | – | 19.0%† |
| METR Time Horizons | METR | – | 74.3% |
| GBAEval | Epoch AI Benchmarking Hub | – | 45.1%† |
| Vending-Bench 2 | Andon Labs | 16484.2† | 6144.2† |
| AutomationBench | Artificial Analysis | 77.8% | – |
| GDP.pdf (AA) | Artificial Analysis | 28.4% | – |
| AA-Briefcase v1.1 | Artificial Analysis | 1609 | – |
| Terminal-Bench-Science 0.1 (AA) | Artificial Analysis | 64.8% | – |
| DeepSearchQA | – | 63.7%† | |
| Maths | |||
| FrontierMath Tiers 1–3 | Epoch AI Benchmarking Hub | – | 78.6% |
| FrontierMath Tier 4 | Epoch AI Benchmarking Hub | – | 49.0% |
| OTIS Mock AIME | Epoch AI Benchmarking Hub | – | 95.3% |
| ProofBench | Vals AI | – | 56.0% |
| LiveBench Mathematics | LiveBench | – | 94.2% |
| AIME (Vals) | Vals AI | – | 96.7% |
| AIME 2026 | MathArena | – | 99.2% |
| HMMT February 2026 | MathArena | – | 97.7% |
| USAMO 2026 | MathArena | – | 95.2% |
| MathArena Apex | MathArena | – | 54.2% |
| Arena Maths | Arena (LMArena) | – | 1460† |
| Knowledge | |||
| SimpleQA Verified | Epoch AI Benchmarking Hub | – | 45.1% |
| LiveBench 1.0 Data Analysis | LiveBench | – | 79.3% |
| AA-Omniscience v1.1 | Artificial Analysis | 53.5 | – |
| MMLU-Pro | Vals AI | – | 87.5% |
| LegalBench | Vals AI | – | 86.0% |
| CorpFin | Vals AI | – | 65.3% |
| TaxEval | Vals AI | – | 74.0% |
| MedQA | Vals AI | – | 96.1% |
| PRBench Finance | Scale AI SEAL | – | 45.6%† |
| PRBench Legal | Scale AI SEAL | – | 44.4%† |
| MultiNRC | Scale AI SEAL | – | 58.3%† |
| CaseLaw v2 | Vals AI | – | 63.8% |
| Excel Modeling Benchmark v1.1 | Vals AI | 76.5% | – |
| MedCode | Vals AI | – | 41.3% |
| MedScribe | Vals AI | – | 77.5% |
| EBR-bench | Epoch AI Benchmarking Hub | – | 25.4% |
| AA-Omniscience v1.1: accuracy | Artificial Analysis | 58.3% | – |
| AA-Omniscience v1.1: non-hallucination | Artificial Analysis | 88.6% | – |
| Instruction following | |||
| LiveBench Language | LiveBench | – | 82.6% |
| LiveBench Instruction Following | LiveBench | – | 70.2% |
| IFBench | Artificial Analysis | – | 74.0% |
| Arena Instruction Following | Arena (LMArena) | – | 1460† |
| Human preference | |||
| Arena Text | Arena (LMArena) | – | 1465† |
| EQ-Bench 4 | EQ-Bench | – | 1258† |
| Arena Creative Writing | Arena (LMArena) | – | 1435† |
| Arena Multi-turn | Arena (LMArena) | – | 1480† |
| Arena Longer Queries | Arena (LMArena) | – | 1477† |
| Multimodal | |||
| Arena Vision | Arena (LMArena) | – | 1292† |
| MMMU-Pro | Artificial Analysis | – | 78.4% |
| VISTA | Scale AI SEAL | – | 50.9%† |
| MMMU-Pro (Vals) | Vals AI | – | 87.5% |
| MortgageTax | Vals AI | – | 68.3% |
| VTB | Scale AI SEAL | – | 29.2%† |
| Blueprint-Bench 2 | Epoch AI Benchmarking Hub | 58.6%† | 27.1%† |
| Long context | |||
| AA-LCR v1.1 | Artificial Analysis | 79.3% | 82.0% |
| Arena Document | Arena (LMArena) | – | 1470† |
| Composite | |||
| Epoch Capabilities Index | Epoch AI Benchmarking Hub | – | 157.3† |
| LiveBench 1.0 | LiveBench | – | 78.0% |
| AA Intelligence Index v5.0 | Artificial Analysis | 59.4 | – |
| AA Intelligence Index v4.3.2 | Artificial Analysis | – | 39.0 |
| AA Progress Index v1 | Artificial Analysis | 351.6 | 311.6 |
| Vals Index | Vals AI | 70.4 | – |
| Vals Finance Index | Vals AI | 66.3% | – |
| Vals RSI Index | Vals AI | – | 15.5% |
| Safety & honesty | |||
| MASK | Scale AI SEAL | – | 89.7%† |
Data as of 2026-10-29. Best configuration of each model; every score links to its source on the model pages.
What changed
- GPT-5.4 (xhigh): 1 new result (AA Progress Index v1).
- GPT-5.4 (xhigh): 2 new results (Harvey's Legal Agent Benchmark v1, Vals RSI Index).
- GPT-5.4 (xhigh): 2 new results (USAMO 2026, τ³-bench Banking).
- GPT-5.4 (xhigh): 1 score revision.
- GPT-5.4 (xhigh): 1 new result (SWE Atlas: Codebase QnA).
- GPT-5.4 (xhigh): 1 score revision.
- GPT-5.4 (xhigh): BenchLeader Index 64.2 → 63.7.
- GPT-5.4 (xhigh): 1 new result (SWE Atlas: Test Writing).
- GPT-5.4 (xhigh): 1 new result (SWE Atlas: Codebase QnA).
Related comparisons
- Gemini 4 Carbon vs Claude Opus 5.5
- Gemini 4 Carbon vs GPT-6 Astra
- Gemini 4 Carbon vs GPT-6.1 Sol
- GPT-5.4 vs Claude Opus 5.5
- GPT-5.4 vs GPT-6 Astra
- GPT-5.4 vs GPT-6.1 Sol
- GPT-5.4 vs Claude Fable 5.1
- Gemini 4 Argon vs GPT-5.4
Rankings: Best LLM overall · Best LLM for coding · Best LLM for agents · Best cheap LLMs
Gemini 4 Carbon vs GPT-5.4: questions
- Is Gemini 4 Carbon better than GPT-5.4?
- Gemini 4 Carbon (high) and GPT-5.4 (xhigh) are neck and neck on quality (65.1 vs 63.3, within the ±6.6 margin). The BenchLeader Index combines every independent quality benchmark; the two are within the margin of error as of 29 Oct 2026, so check the category scores for your use.
- Gemini 4 Carbon vs GPT-5.4: which is cheaper, and will that change?
- GPT-5.4 is cheaper today: $5.63 against $8.00 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
- Is Gemini 4 Carbon or GPT-5.4 better for coding?
- Gemini 4 Carbon scores higher in coding (72 vs 65 on the category index, where 50 is average). On SciCode, the top shared coding benchmark, it is 64.2% for Gemini 4 Carbon against 56.6% for GPT-5.4.
- Which is faster, Gemini 4 Carbon or GPT-5.4?
- GPT-5.4 streams faster (102 vs 100 output tokens per second, Artificial Analysis 7-day median).
- Which is better for long documents?
- GPT-5.4 scores higher on long-context benchmarks (66 vs 65). Both take about 1.1M tokens: Gemini 4 Carbon 1,000,000, GPT-5.4 1,050,000.
More questions (3)
- Which has lower latency?
- Gemini 4 Carbon answers first: 103.1 s against 148.2 s to the first answer token, as Artificial Analysis measures it. Reasoning effort drives most of this: a higher setting thinks longer before it answers.
- Do independent benchmarks agree on Gemini 4 Carbon vs GPT-5.4?
- Across the 6 benchmarks both have, Gemini 4 Carbon leads 5, GPT-5.4 leads 1.
- When was Gemini 4 Carbon released?
- Gemini 4 Carbon was released on 28 Oct 2026.