Gemini 4 Carbon vs GPT-5.6 Luna
Full profiles: Gemini 4 Carbon (Google) and GPT-5.6 Luna (OpenAI): every benchmark, price and rank.
Updated · Data from 7 sources
As of 29 Oct 2026, Gemini 4 Carbon is ahead of GPT-5.6 Luna on quality (65.1 vs 58.5 on the BenchLeader Index), but GPT-5.6 Luna is 18× cheaper at $0.450 vs $8.00 per 1M tokens blended. Choose Gemini 4 Carbon for reasoning (79 vs 53) and coding (72 vs 57); choose GPT-5.6 Luna for fast first answers (35.8 s vs 103.1 s) and long context (66 vs 65), where it also streams faster.
Key facts (as of 29 Oct 2026)
- Gemini 4 Carbon scores 65.1 on the BenchLeader Index and GPT-5.6 Luna scores 58.5 (BenchLeader, 29 Oct 2026).
- Gemini 4 Carbon costs $8.00 per 1M tokens blended (list price, 29 Oct 2026).
- GPT-5.6 Luna costs $0.450 per 1M tokens blended (list price, 29 Oct 2026).
- GPT-5.6 Luna streams faster than Gemini 4 Carbon (Artificial Analysis, 7-day median, 29 Oct 2026).
- Of the 8 benchmarks both models have, Gemini 4 Carbon leads 7 (BenchLeader, 29 Oct 2026).
Gemini 4 Carbon benchmarks high Google
GPT-5.6 Luna benchmarks xhigh OpenAIOpen in the comparison toolWhich 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 57 category score; SciCode 64.2% vs 50.5% | High |
| Reasoning | Gemini 4 Carbon | 79 vs 53 category score; Humanity's Last Exam (AA) 60.4% vs 37.0% | High |
| Long documents | GPT-5.6 Luna | 66 vs 65 category score; AA-LCR v1.1 81.7% vs 79.3% | Medium |
| Chat quality (human preference) | – | Only GPT-5.6 Luna has an Arena rating (64 category score) | Not comparable |
| Lowest cost | GPT-5.6 Luna | $0.450 vs $8.00 per 1M blended; agentic coding session $0.017 vs $0.320 | – |
| Fastest first answer | GPT-5.6 Luna | 35.8 s (GPT-5.6 Luna (xhigh)) vs 103.1 s (Gemini 4 Carbon (high)) 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) leads on quality: 65.1 vs 58.5.
- Gemini 4 Carbon (high) is stronger in coding, knowledge, reasoning.
- GPT-5.6 Luna (xhigh) is stronger in long context, agents & tools, human preference, maths, multimodal.
- GPT-5.6 Luna (xhigh) is 18× cheaper ($0.450 vs $8.00 per 1M blended).
- GPT-5.6 Luna (xhigh) streams 1.3× faster (131 vs 100 tokens per second).
At a glance
| Spec | Gemini 4 Carbon | GPT-5.6 Luna |
|---|---|---|
| Developer | OpenAI | |
| Released | 28 Oct 2026 | 9 Jul 2026 |
| Context window (tokens) | 1,000,000 | 1,050,000 |
| Compared at | Gemini 4 Carbon (high) | GPT-5.6 Luna (xhigh) |
| Open weights | No | No |
| Price in / out / cached input per 1M | $4.00 / $20.00 / – | $0.200 / $1.20 / $0.020 |
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-5.6 Luna (xhigh) | Leader |
|---|---|---|---|
| Reasoning | 78.9 | 52.9 | Gemini 4 Carbon |
| Coding | 71.7 | 57.0 | Gemini 4 Carbon |
| Agents & tools | – | 57.1 | not comparable |
| Maths | – | 65.9 | not comparable |
| Knowledge | 75.8 | – | not comparable |
| Human preference | – | 63.8 | not comparable |
| Multimodal | – | 60.1 | not comparable |
| Long context | 64.5 | 65.7 | GPT-5.6 Luna |
Where sources disagree
Across the 8 benchmarks both have, Gemini 4 Carbon leads 7, GPT-5.6 Luna leads 1. Every overall index that rates both puts Gemini 4 Carbon ahead.
| Overall index | Gemini 4 Carbon (high) | GPT-5.6 Luna (xhigh) | Leader |
|---|---|---|---|
| BenchLeader Index | 65.1 | 58.5 | Gemini 4 Carbon |
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-5.6 Luna (xhigh) |
|---|---|---|
| Blended price per 1M tokens | $8.00 | $0.450 |
| Input / output per 1M | $4.00 / $20.00 | $0.200 / $1.20 |
| Cached input per 1M | – | $0.020 |
| Price per task: chat reply | $0.0076 | $0.0004 |
| Price per task: summarize a 30-page report | $0.060 | $0.0031 |
| Price per task: code edit | $0.054 | $0.0030 |
| Price per task: agentic coding session | $0.320 | $0.017 |
| Price per task: structured extraction | $0.012 | $0.0006 |
| Cost to run a full benchmark suite | $7.76 | – |
| Output speed | 100 tok/s | 131 tok/s |
| Time to first answer | 103.1 s | 35.8 s |
| Time to a complete answer | 108.1 s | 39.6 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-5.6 Luna (xhigh) scores 58.5 and takes 36 s to its first answer token; GPT-5.6 Luna (max) scores 57.8 in 104 s, GPT-5.6 Luna (high) scores 54.9 in 8 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
113 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-5.6 Luna (xhigh) |
|---|---|---|---|
| BenchLeader Index | BenchLeader | 65.1 | 58.5 |
| Coding score | BenchLeader | 71.7 | 57.0 |
| Knowledge score | BenchLeader | 75.8 | – |
| Long context score | BenchLeader | 64.5 | 65.7 |
| Reasoning score | BenchLeader | 78.9 | 52.9 |
| Agents & tools score | BenchLeader | – | 57.1 |
| Human preference score | BenchLeader | – | 63.8 |
| Maths score | BenchLeader | – | 65.9 |
| Multimodal score | BenchLeader | – | 60.1 |
| Blended price $/M | BenchLeader | $8.00 | $0.450 |
| Output speed | BenchLeader | 100 tok/s | 131 tok/s |
| Time to first answer | BenchLeader | 103.1 s | 35.8 s |
| Context window | BenchLeader | 1,000,000 tokens | 1,050,000 tokens |
| Reasoning | |||
| GPQA Diamond | Epoch AI Benchmarking Hub | – | 91.6%† |
| SimpleBench | SimpleBench | – | 46.8% |
| Arena Hard Prompts | Arena (LMArena) | – | 1475 |
| LiveBench 1.0 Reasoning | LiveBench | – | 85.6%† |
| GPQA Diamond (AA) | Artificial Analysis | – | 89.5% |
| Humanity's Last Exam (AA) | Artificial Analysis | 60.4% | 37.0% |
| HLE-Diamond (AA) | Artificial Analysis | 69.1% | – |
| GPQA Diamond (Vals) | Vals AI | – | 91.7%† |
| Kagi LLM Benchmark | Kagi LLM Benchmark | – | 49.1%† |
| ARC-AGI-1 | ARC Prize | – | 87.7% |
| ARC-AGI-2 | ARC Prize | – | 47.6% |
| ARC-AGI-3 | ARC Prize | – | 0.0% |
| CritPt | Artificial Analysis | 36.0% | 20.6% |
| CritPt v2 hard subset (AA) | Artificial Analysis | 63.0% | – |
| MysteryMechanism | Vals AI | – | 14.4%† |
| Chess Puzzles | Epoch AI Benchmarking Hub | – | 43.0%† |
| Mystery Game Puzzles | Epoch AI Benchmarking Hub | – | 21.0%† |
| LMCA | Epoch AI Benchmarking Hub | – | 48.5% |
| DTBench | Epoch AI Benchmarking Hub | – | 89.1% |
| Coding | |||
| SciCode | SciCode | 64.2% | 50.5% |
| WeirdML | WeirdML | – | 60.9%† |
| FrontierCode | Cognition | – | 39.8%† |
| Arena Coding | Arena (LMArena) | – | 1502 |
| Arena WebDev | Arena (LMArena) | – | 1519 |
| LiveBench 1.0 Coding | LiveBench | – | 82.9%† |
Show all 113 benchmarks
| Metric | Publisher | Gemini 4 Carbon (high) | GPT-5.6 Luna (xhigh) |
|---|---|---|---|
| Coding | |||
| IOI | Vals AI | – | 61.8%† |
| SWE-bench (Vals) | Vals AI | – | 93.0%† |
| Code Migration | Vals AI | – | 44.5%† |
| ProgramBench | Vals AI | – | 0.0%† |
| Vibe Code Bench 1-100 | Vals AI | – | 22.6%† |
| Vibe Code Bench v1.1 | Vals AI | 89.1% | 77.1%† |
| Surface Evolver Bench | Epoch AI Benchmarking Hub | – | 61.9%† |
| DeepSWE v1.1 | Epoch AI Benchmarking Hub | – | 56.9% |
| CursorBench | Cursor | – | 33.0% |
| ALE-Bench | Epoch AI Benchmarking Hub | – | 1667.4† |
| Terminal-Bench 4.0 (AA) | Artificial Analysis | 65.2% | 3.5% |
| Terminal-Bench 2.1 (AA) | Artificial Analysis | – | 77.9% |
| Startrise Frontend | Startrise | – | 65.0%† |
| FrontierCode 1.1 (AA) | Artificial Analysis | 48.5% | – |
| FrontierCode 2.0 (AA) | Artificial Analysis | 40.6% | – |
| Agents & tools | |||
| Terminal-Bench | Terminal-Bench | – | 17.3%† |
| APEX-Agents | Mercor | – | 43.0%† |
| LiveBench Agentic Coding | LiveBench | – | 48.4%† |
| Terminal-Bench 2.1 (Vals) | Vals AI | – | 79.0%† |
| τ³-Banking (AA) | Artificial Analysis | – | 28.7% |
| ITBench SRE (AA) | Artificial Analysis | – | 40.3%† |
| APEX-Agents (AA) | Artificial Analysis | – | 35.8%† |
| CyberBench | Vals AI | – | 73.6%† |
| Finance Agent v2 | Vals AI | – | 55.0%† |
| 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 | – | 1.3%† |
| Legal Research Bench | Vals AI | 55.8% | 36.5%† |
| SkillsBench | Vals AI | – | 60.5%† |
| SREBench | Vals AI | – | 1.5%† |
| Tax Agent Bench | Vals AI | 48.9% | 24.8%† |
| Terminal-Bench 4.0 (Vals) | Vals AI | 63.1% | 11.6%† |
| Terminal-Bench Science | Vals AI | – | 0.0%† |
| BALROG | BALROG | – | 45.6%† |
| Vending-Bench 2 | Andon Labs | 16484.2† | 4094.7† |
| AutomationBench | Artificial Analysis | 77.8% | 42.6% |
| GDP.pdf (AA) | Artificial Analysis | 28.4% | 23.8% |
| EnterpriseOps-Gym | Artificial Analysis | – | 40.8%† |
| AA-Briefcase v1.1 | Artificial Analysis | 1609 | 1339† |
| Terminal-Bench-Science 0.1 (AA) | Artificial Analysis | 64.8% | 3.3%† |
| FACTS Search | – | 79.3%† | |
| Maths | |||
| FrontierMath Tiers 1–3 | Epoch AI Benchmarking Hub | – | 82.1%† |
| FrontierMath Tier 4 | Epoch AI Benchmarking Hub | – | 61.0%† |
| OTIS Mock AIME | Epoch AI Benchmarking Hub | – | 98.3%† |
| ProofBench | Vals AI | – | 60.0%† |
| LiveBench Mathematics | LiveBench | – | 87.2%† |
| Arena Maths | Arena (LMArena) | – | 1478 |
| Knowledge | |||
| SimpleQA Verified | Epoch AI Benchmarking Hub | – | 41.0%† |
| LiveBench 1.0 Data Analysis | LiveBench | – | 78.0%† |
| AA-Omniscience v1.1 | Artificial Analysis | 53.5 | – |
| MMLU-Pro | Vals AI | – | 86.0%† |
| LegalBench | Vals AI | – | 84.0%† |
| CorpFin | Vals AI | – | 64.2%† |
| TaxEval | Vals AI | – | 76.2%† |
| BioMysteryBench | Vals AI | – | 61.5%† |
| Excel Modeling Benchmark | Vals AI | – | 67.1%† |
| Excel Modeling Benchmark v1.1 | Vals AI | 76.5% | – |
| MedCode | Vals AI | – | 42.4%† |
| MedScribe | Vals AI | – | 84.4%† |
| Public Benefits Bench v1.1 | Vals AI | – | 61.2%† |
| GDP.pdf | Epoch AI Benchmarking Hub | – | 22.7%† |
| AA-Omniscience v1.1: accuracy | Artificial Analysis | 58.3% | – |
| AA-Omniscience v1.1: non-hallucination | Artificial Analysis | 88.6% | – |
| Instruction following | |||
| LiveBench Language | LiveBench | – | 72.6%† |
| LiveBench Instruction Following | LiveBench | – | 60.1%† |
| Arena Instruction Following | Arena (LMArena) | – | 1444 |
| Human preference | |||
| Arena Text | Arena (LMArena) | – | 1453 |
| EQ-Bench 4 | EQ-Bench | – | 1146† |
| Arena Creative Writing | Arena (LMArena) | – | 1410 |
| Arena Multi-turn | Arena (LMArena) | – | 1459 |
| Arena Longer Queries | Arena (LMArena) | – | 1452 |
| Multimodal | |||
| Arena Vision | Arena (LMArena) | – | 1257 |
| MMMU-Pro | Artificial Analysis | – | 78.5% |
| MMMU-Pro (Vals) | Vals AI | – | 85.0%† |
| MortgageTax | Vals AI | – | 67.3%† |
| Blueprint-Bench 2 | Epoch AI Benchmarking Hub | 58.6%† | 22.6%† |
| Long context | |||
| AA-LCR v1.1 | Artificial Analysis | 79.3% | 81.7% |
| Arena Document | Arena (LMArena) | – | 1462 |
| MLCR | Artificial Analysis | – | 15.0% |
| Composite | |||
| Epoch Capabilities Index | Epoch AI Benchmarking Hub | – | 156.5† |
| LiveBench 1.0 | LiveBench | – | 73.6%† |
| AA Intelligence Index v5.0 | Artificial Analysis | 59.4 | – |
| AA Intelligence Index v4.3.2 | Artificial Analysis | – | 34.6 |
| AA Progress Index v1 | Artificial Analysis | 351.6 | 305.8 |
| Vals Index | Vals AI | 70.4 | – |
| Vals Finance Index | Vals AI | 66.3% | – |
| Vals Science Index | Vals AI | – | 25.3%† |
| Vals Multimodal Index | Vals AI | – | 69.1%† |
Data as of 2026-10-29. Best configuration of each model; every score links to its source on the model pages.
What changed
- GPT-5.6 Luna (xhigh): 1 new result (AA Progress Index v1).
- GPT-5.6 Luna (xhigh): 1 score revision.
- GPT-5.6 Luna (xhigh): 3 new results (AutomationBench, GDP.pdf (AA), MLCR).
- GPT-5.6 Luna (xhigh): 1 score revision.
- GPT-5.6 Luna (xhigh): 1 score revision.
- GPT-5.6 Luna (xhigh): 1 score revision.
- GPT-5.6 Luna (xhigh): BenchLeader Index 58.8 → 58.3.
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-5.6 Luna vs Claude Opus 5.5
- GPT-5.6 Luna vs GPT-6 Astra
- GPT-5.6 Luna vs GPT-6.1 Sol
- GPT-5.6 Luna vs Claude Fable 5.1
- Gemini 4 Argon vs GPT-5.6 Luna
Rankings: Best LLM overall · Best LLM for coding · Best LLM for agents · Best cheap LLMs
Gemini 4 Carbon vs GPT-5.6 Luna: questions
- Is Gemini 4 Carbon better than GPT-5.6 Luna?
- Gemini 4 Carbon (high) leads on quality: 65.1 vs 58.5. The BenchLeader Index combines every independent quality benchmark; Gemini 4 Carbon (high) is ahead overall as of 29 Oct 2026, but check the category scores for your use.
- Gemini 4 Carbon vs GPT-5.6 Luna: which is cheaper, and will that change?
- GPT-5.6 Luna is cheaper today: $0.450 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-5.6 Luna better for coding?
- Gemini 4 Carbon scores higher in coding (72 vs 57 on the category index, where 50 is average). On SciCode, the top shared coding benchmark, it is 64.2% for Gemini 4 Carbon against 50.5% for GPT-5.6 Luna.
- Which is faster, Gemini 4 Carbon or GPT-5.6 Luna?
- GPT-5.6 Luna streams faster (131 vs 100 output tokens per second, Artificial Analysis 7-day median).
- Which is better for long documents?
- GPT-5.6 Luna scores higher on long-context benchmarks (66 vs 65). Both take about 1.1M tokens: Gemini 4 Carbon 1,000,000, GPT-5.6 Luna 1,050,000.
More questions (3)
- Which has lower latency?
- GPT-5.6 Luna answers first: 35.8 s against 103.1 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-5.6 Luna?
- Across the 8 benchmarks both have, Gemini 4 Carbon leads 7, GPT-5.6 Luna leads 1. Every overall index that rates both puts Gemini 4 Carbon ahead.
- When was Gemini 4 Carbon released?
- Gemini 4 Carbon was released on 28 Oct 2026.