Gemini 4 Carbon vs Gemini 4 Neon
Full profiles: Gemini 4 Carbon (Google) and Gemini 4 Neon (Google): every benchmark, price and rank.
Updated · Data from 12 sources
As of 19 Nov 2026, Gemini 4 Carbon and Gemini 4 Neon are neck and neck on quality (68.0 vs 67.3 on the BenchLeader Index, inside the ±3.3 margin), and Gemini 4 Carbon is 1.5× cheaper at $4.00 vs $6.00 per 1M tokens blended. Choose 4 Carbon for knowledge (75 vs 72); choose 4 Neon for fast first answers (40.6 s vs 72.4 s) and agents & tools (76 vs 69), where it also streams ~1.5× faster.
Key facts (as of 19 Nov 2026)
- Gemini 4 Carbon scores 68.0 on the BenchLeader Index and Gemini 4 Neon scores 67.3 (BenchLeader, 19 Nov 2026).
- Gemini 4 Carbon costs $4.00 per 1M tokens blended (list price, 19 Nov 2026).
- Gemini 4 Neon costs $6.00 per 1M tokens blended (list price, 19 Nov 2026).
- Gemini 4 Neon streams ~1.5× faster than Gemini 4 Carbon (Artificial Analysis, 7-day median, 19 Nov 2026).
- Of the 23 benchmarks both models have, Gemini 4 Neon leads 12 (BenchLeader, 19 Nov 2026).
- On Arena WebDev, Gemini 4 Carbon is rated 1769 and Gemini 4 Neon 1764 (LMArena, as published by 19 Nov 2026).
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 | Either | 73 vs 73 category score; Arena WebDev 1769 vs 1764 | Either |
| Agents & tool use | Gemini 4 Neon | 76 vs 69 category score; Arena Agent 18.4 vs 16.1 | Medium |
| Reasoning | Either | 72 vs 72 category score; HLE-Diamond (AA) 69.1% vs 68.4% | Either |
| Knowledge | Gemini 4 Carbon | 75 vs 72 category score; AA-Omniscience v1.1 53.5 vs 53.8 | Medium |
| Chat quality (human preference) | Either | 73 vs 72 on Arena votes | Either |
| Lowest cost | Gemini 4 Neon | $6.00 vs $8.00 per 1M blended; agentic coding session $0.240 vs $0.320 | – |
| Fastest first answer | Gemini 4 Neon | 40.6 s (Gemini 4 Neon (high)) vs 72.4 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) and Gemini 4 Neon (high) are neck and neck on quality (68.0 vs 67.3, within the ±3.3 margin).
- Gemini 4 Carbon (high) is stronger in human preference, knowledge.
- Gemini 4 Neon (high) is stronger in agents & tools, coding, reasoning.
- Gemini 4 Carbon (high) is 1.5× cheaper ($4.00 vs $6.00 per 1M blended).
- Gemini 4 Neon (high) streams 1.9× faster (101 vs 55 tokens per second).
At a glance
| Spec | Gemini 4 Carbon | Gemini 4 Neon |
|---|---|---|
| Developer | ||
| Released | 28 Oct 2026 | 18 Nov 2026 |
| Context window (tokens) | 1,048,576 | 1,048,576 |
| Compared at | Gemini 4 Carbon (high) | Gemini 4 Neon (high) |
| Open weights | No | No |
| Price in / out / cached input per 1M | $4.00 / $20.00 / $0.100 | $3.00 / $15.00 / $0.150 |
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) | Gemini 4 Neon (high) | Leader |
|---|---|---|---|
| Reasoning | 71.8 | 71.9 | level |
| Coding | 72.8 | 73.4 | level |
| Agents & tools | 69.1 | 76.2 | Gemini 4 Neon |
| Maths | 71.3 | – | not comparable |
| Knowledge | 74.7 | 72.3 | Gemini 4 Carbon |
| Human preference | 72.6 | 71.7 | level |
| Long context | 64.3 | – | not comparable |
| Composite | 67.8 | – | not comparable |
On one chart
Each domain on the index's own scale, where 50 is an average model. Click a domain to see the benchmarks behind it. Pick up to 5 models from the list.
Not drawn, measured for fewer than two of these models: Maths, Instruction following, Long context, Multimodal.
Where sources disagree
Across the 23 benchmarks both have, Gemini 4 Carbon leads 11, Gemini 4 Neon leads 12. Every overall index that rates both puts Gemini 4 Carbon ahead.
| Overall index | Gemini 4 Carbon (high) | Gemini 4 Neon (high) | Leader |
|---|---|---|---|
| BenchLeader Index | 68.0 | 67.3 | neck and neck |
| AA Intelligence Index v5.0.1 | 60.0 | 59.6 | Gemini 4 Carbon |
| Arena Text | 1526 | 1519 | 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. Cost, tokens and time per benchmark task are measured: Artificial Analysis’ averages over its whole Intelligence Index run.
| Measure | Gemini 4 Carbon (high) | Gemini 4 Neon (high) |
|---|---|---|
| Blended price per 1M tokens | $8.00 | $6.00 |
| Input / output per 1M | $4.00 / $20.00 | $3.00 / $15.00 |
| Cached input per 1M | $0.100 | $0.150 |
| Price per task: chat reply | $0.0076 | $0.0057 |
| Price per task: summarize a 30-page report | $0.060 | $0.045 |
| Price per task: code edit | $0.054 | $0.041 |
| Price per task: agentic coding session | $0.320 | $0.240 |
| Price per task: structured extraction | $0.012 | $0.0090 |
| Agentic coding session, input mostly cached | $0.144 | $0.112 |
| Cost per benchmark task (AA Index) | $7.76 | $3.74 |
| Output tokens per task (AA Index) | 123k | 90k |
| Of which reasoning | 72k | 46k |
| Time per task (AA Index) | 38 min | 15 min |
| Output speed | 55 tok/s | 101 tok/s |
| Time to first answer | 72.4 s | 40.6 s |
| Time to a complete answer, last measured 13 Nov 2026 | 160.6 s | – |
Per token is not per task: an average task of Artificial Analysis' Intelligence Index cost $7.76 on Gemini 4 Carbon (high) and $3.74 on Gemini 4 Neon (high), and Gemini 4 Carbon (high) wrote 1.4× as many output tokens for the same work, which a per-token price hides.
Each model is compared at its best-ranked reasoning-effort setting. Gemini 4 Carbon (high) scores 68.0 and takes 72 s to its first answer token; Gemini 4 Carbon (medium) scores 64.7 in 33 s, Gemini 4 Carbon (low) scores 59.0 in 9 s. Gemini 4 Neon (high) scores 67.3 and takes 41 s to its first answer token; Gemini 4 Neon (xhigh) scores 64.3 in 57 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
94 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) | Gemini 4 Neon (high) |
|---|---|---|---|
| BenchLeader Index | BenchLeader | 68.0 | 67.3 |
| Agents & tools score | BenchLeader | 69.1 | 76.2 |
| Coding score | BenchLeader | 72.8 | 73.4 |
| Composite score | BenchLeader | 67.8 | – |
| Human preference score | BenchLeader | 72.6 | 71.7 |
| Knowledge score | BenchLeader | 74.7 | 72.3 |
| Long context score | BenchLeader | 64.3 | – |
| Maths score | BenchLeader | 71.3 | – |
| Reasoning score | BenchLeader | 71.8 | 71.9 |
| Blended price $/M | BenchLeader | $8.00 | $6.00 |
| Output speed | BenchLeader | 55 tok/s | 101 tok/s |
| Time to first answer | BenchLeader | 72.4 s | 40.6 s |
| Context window | BenchLeader | 1,048,576 tokens | 1,048,576 tokens |
| Reasoning | |||
| GPQA Diamond | Epoch AI Benchmarking Hub | 96.5% | – |
| SimpleBench | SimpleBench | 92.6%† | – |
| Arena Hard Prompts | Arena (LMArena) | 1546 | 1546 |
| LiveBench 2.0 Reasoning | LiveBench | 90.6% | 89.6%† |
| HLE-Diamond (AA) | Artificial Analysis | 69.1% | 68.4% |
| ARC-AGI-1 | ARC Prize | 97.5% | – |
| ARC-AGI-2 | ARC Prize | 92.1% | – |
| ARC-AGI-3 | ARC Prize | 96.9% | – |
| CritPt v2 hard subset (AA) | Artificial Analysis | 69.0% | 59.0% |
| Arena Expert | Arena (LMArena) | 1566 | 1541 |
| Chess Puzzles | Epoch AI Benchmarking Hub | 78.0% | – |
| Mystery Game Puzzles | Epoch AI Benchmarking Hub | 73.0% | – |
| Coding | |||
| SciCode | SciCode | 64.2% | – |
| Arena Coding | Arena (LMArena) | 1567 | 1558 |
| Arena WebDev | Arena (LMArena) | 1769 | 1764 |
| LiveBench 2.0 Coding | LiveBench | 91.0% | 89.7%† |
| Vibe Code Bench v1.1 | Vals AI | 89.1% | – |
| Vibe Code Bench 1-100 v1.1 | Vals AI | 49.7% | – |
| SWE Atlas v1.1 | Scale AI SEAL | 71.5%† | – |
| SWE Atlas v1.1: Codebase QnA | Scale AI SEAL | 73.4%† | – |
| SWE Atlas v1.1: Test Writing | Scale AI SEAL | 76.7%† | – |
| SWE Atlas v1.1: Refactoring | Scale AI SEAL | 64.3%† | – |
| CursorBench | Cursor | 60.4% | – |
| ALE-Bench | Epoch AI Benchmarking Hub | 2374.7 | – |
| FrontierSWE | Epoch AI Benchmarking Hub | 61.2% | – |
Show all 94 benchmarks
| Metric | Publisher | Gemini 4 Carbon (high) | Gemini 4 Neon (high) |
|---|---|---|---|
| Coding | |||
| Terminal-Bench 4.0 (AA) | Artificial Analysis | 65.2% | 61.6% |
| Design Arena Website | Design Arena | 1304† | – |
| FrontierCode 2.0 (AA) | Artificial Analysis | 40.6% | 43.3% |
| Agents & tools | |||
| Terminal-Bench | Terminal-Bench | 64.5% | – |
| APEX-Agents | Mercor | 84.2%† | – |
| Arena Agent | Arena (LMArena) | 16.1 | 18.4 |
| LiveBench Long Agentic | LiveBench | 71.8% | 82.6%† |
| LiveBench Browser | LiveBench | 80.0% | 95.6%† |
| HiL-Bench | Scale AI SEAL | 50.3%† | – |
| GDPval-AA v2.1 | Artificial Analysis | 1733 | – |
| Finance Agent v2.1 | Vals AI | 67.2% | – |
| Harvey's Legal Agent Benchmark v1.1 | Vals AI | 17.5% | – |
| Legal Research Bench | Vals AI | 55.8% | – |
| Tax Agent Bench | Vals AI | 51.8% | – |
| Terminal-Bench 4.0 (Vals) | Vals AI | 63.1% | – |
| Arena Agent Task Completion | Arena (LMArena) | 17.2 | 19.2 |
| PostTrainBench | Epoch AI Benchmarking Hub | 41.0% | – |
| Vending-Bench 2 | Andon Labs | 16484.2† | – |
| AutomationBench | Artificial Analysis | 77.8% | 79.2% |
| GDP.pdf (AA) | Artificial Analysis | 28.4% | 33.4% |
| AA-Briefcase v1.1 | Artificial Analysis | 1611 | 1632 |
| GDPx | GDPx by SemiAnalysis | 10.9% | – |
| Design Arena Agentic | Design Arena | 1304† | – |
| Terminal-Bench-Science 0.1 (AA) | Artificial Analysis | 64.8% | 65.7% |
| MKA (AA) | Artificial Analysis | 54.3% | 54.6% |
| FACTS Search | 94.4%† | – | |
| Maths | |||
| FrontierMath Tiers 1–3 | Epoch AI Benchmarking Hub | 90.5% | – |
| FrontierMath Tier 4 | Epoch AI Benchmarking Hub | 95.1% | – |
| OTIS Mock AIME | Epoch AI Benchmarking Hub | 100.0% | – |
| ArXivMath | MathArena | 94.3% | – |
| BrokenArXiv | MathArena | 84.4% | – |
| Autoform-Hard | Vals AI | 71.4% | – |
| FrontierMath Erdős | Epoch AI Benchmarking Hub | 2.9% | – |
| Knowledge | |||
| SimpleQA Verified | Epoch AI Benchmarking Hub | 77.7% | – |
| LiveBench 2.0 Data Analysis | LiveBench | 86.5% | 85.3%† |
| LiveBench Research | LiveBench | 92.5% | 94.5%† |
| AA-Omniscience v1.1 | Artificial Analysis | 53.5 | 53.8 |
| PRBench Legal | Scale AI SEAL | 53.4%† | – |
| Excel Modeling Benchmark v1.1 | Vals AI | 76.5% | – |
| DrugDiscoveryBench | Scale AI SEAL | 71.5% | – |
| EBR-bench | Epoch AI Benchmarking Hub | 87.6% | – |
| GDP.pdf | Epoch AI Benchmarking Hub | 56.4% | – |
| AA-Omniscience v1.1: accuracy | Artificial Analysis | 58.3% | 59.4% |
| AA-Omniscience v1.1: non-hallucination | Artificial Analysis | 88.6% | 86.1% |
| FACTS Parametric | 81.6%† | – | |
| Instruction following | |||
| Arena Instruction Following | Arena (LMArena) | 1524 | 1522 |
| Human preference | |||
| Arena Text | Arena (LMArena) | 1526 | 1519 |
| Arena Creative Writing | Arena (LMArena) | 1515 | 1514 |
| Arena Multi-turn | Arena (LMArena) | 1563 | 1529 |
| Arena Longer Queries | Arena (LMArena) | 1537 | 1521 |
| Multimodal | |||
| LiveBench Design | LiveBench | 80.2% | 81.5%† |
| MMMU-Pro | Artificial Analysis | 89.3%† | – |
| Humanity's Sixth Sense | Scale AI SEAL | 71.1% | 75.9%† |
| Blueprint-Bench 2 | Epoch AI Benchmarking Hub | 58.6%† | – |
| Furniture Assembly | Epoch AI Benchmarking Hub | 77.5% | – |
| VIALS-AA | Handshake AI | 64.0% | – |
| FACTS Multimodal | 60.2%† | – | |
| Long context | |||
| AA-LCR v1.1 | Artificial Analysis | 79.3% | – |
| MLCR | Artificial Analysis | 19.4% | 28.9% |
| FACTS Grounding | 77.2%† | – | |
| Composite | |||
| Epoch Capabilities Index | Epoch AI Benchmarking Hub | 169.8† | – |
| LiveBench 2.0 | LiveBench | 83.3% | 87.4%† |
| AA Intelligence Index v5.0.1 | Artificial Analysis | 60.0 | 59.6 |
| AA Progress Index v1 | Artificial Analysis | 350.8 | – |
| Vals Index | Vals AI | 70.5 | – |
| Vals Finance Index | Vals AI | 67.0% | – |
| Vals RSI Index | Vals AI | 27.7% | – |
| Vals Legal Index | Vals AI | 75.9% | – |
| Safety & honesty | |||
| FORTRESS | Scale AI SEAL | 16.4%† | – |
Data as of 2026-11-19. Best configuration of each model; every score links to its source on the model pages.
What changed
- Gemini 4 Carbon (high): 2 score revisions; input price moved from $2.00/M to $4.00/M (100% up); output price moved from $10.00/M to $20.00/M (100% up); cost per task moved from $4.90/M to $7.76/M (58% up).
- Gemini 4 Neon (high): 13 new results (MLCR, MKA (AA), Arena Text…).
- Gemini 4 Carbon (high): cost per task moved from $5.48/M to $4.90/M (11% down).
- Gemini 4 Neon (high): first results.
- Gemini 4 Carbon (high): BenchLeader Index 68.9 → 68.3.
- Gemini 4 Carbon (high): 3 new results (PostTrainBench, ALE-Bench, GDP.pdf).
- Gemini 4 Carbon (high): 2 score revisions.
- Gemini 4 Carbon (high): 8 new results (LiveBench 2.0, LiveBench Long Agentic, LiveBench 2.0 Reasoning…).
- Gemini 4 Carbon (high): BenchLeader Index 69.9 → 68.9.
- Gemini 4 Carbon (high): 1 new result (Terminal-Bench).
Related comparisons
- Gemini 4 Carbon vs Claude Fable 5.2
- Gemini 4 Carbon vs GPT-6.2 Astra
- Gemini 4 Carbon vs Claude Opus 5.5
- Gemini 4 Neon vs Claude Fable 5.2
- Gemini 4 Neon vs GPT-6.2 Astra
- Gemini 4 Neon vs Claude Opus 5.5
- Gemini 4 Argon vs Gemini 4 Neon
- Gemini 4 Carbon vs Gemini 4 Hydrogen
Rankings: Best LLM overall · Best LLM for coding · Best LLM for agents · Best cheap LLMs
Gemini 4 Carbon vs Gemini 4 Neon: questions
- Is Gemini 4 Carbon better than Gemini 4 Neon?
- Gemini 4 Carbon (high) and Gemini 4 Neon (high) are neck and neck on quality (68.0 vs 67.3, within the ±3.3 margin). The BenchLeader Index combines every independent quality benchmark; the two are within the margin of error as of 19 Nov 2026, so check the category scores for your use.
- Gemini 4 Carbon vs Gemini 4 Neon: which is cheaper, and will that change?
- Gemini 4 Carbon is cheaper today: $4.00 against $6.00 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
- Is Gemini 4 Carbon or Gemini 4 Neon better for coding?
- They are level in coding (73 vs 73 on the category index). On Arena WebDev, the top shared coding benchmark, it is 1769 for Gemini 4 Carbon against 1764 for Gemini 4 Neon.
- Which is faster, Gemini 4 Carbon or Gemini 4 Neon?
- Gemini 4 Neon streams ~1.5× faster (101 vs 55 output tokens per second, Artificial Analysis 7-day median).
- Is Gemini 4 Carbon or Gemini 4 Neon better for agents and tool use?
- Gemini 4 Neon scores higher in agentic tasks (76 vs 69 on the category index, where 50 is average). On Arena Agent, the top shared agents & tools benchmark, it is 16.1 for Gemini 4 Carbon against 18.4 for Gemini 4 Neon.
More questions (4)
- Which is better for long documents?
- Both accept 1,048,576 tokens of context.
- Which has lower latency?
- Gemini 4 Neon answers first: 40.6 s against 72.4 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 Gemini 4 Neon?
- Across the 23 benchmarks both have, Gemini 4 Carbon leads 11, Gemini 4 Neon leads 12. Every overall index that rates both puts Gemini 4 Carbon ahead.
- When was Gemini 4 Neon released?
- Gemini 4 Neon was released on 18 Nov 2026.