Gemini 3.7 Flash vs GPT-5.3-Codex
Full profiles: Gemini 3.7 Flash (Google) and GPT-5.3-Codex (OpenAI): every benchmark, price and rank.
Updated · Data from 10 sources
As of 16 Nov 2026, Gemini 3.7 Flash and GPT-5.3-Codex are neck and neck on quality (62.1 vs 62.9 on the BenchLeader Index, inside the ±7.8 margin), and Gemini 3.7 Flash is 3.2× cheaper at $1.50 vs $4.81 per 1M tokens blended. Choose Gemini 3.7 Flash for multimodal (67 vs 61) and coding (62 vs 61) and fast first answers (2.2 s vs 38.2 s); choose GPT-5.3-Codex for agents & tools (80 vs 49).
Key facts (as of 16 Nov 2026)
- Gemini 3.7 Flash scores 62.1 on the BenchLeader Index and GPT-5.3-Codex scores 62.9 (BenchLeader, 16 Nov 2026).
- Gemini 3.7 Flash costs $1.50 per 1M tokens blended (list price, 16 Nov 2026).
- GPT-5.3-Codex costs $4.81 per 1M tokens blended (list price, 16 Nov 2026).
- Of the 12 benchmarks both models have, Gemini 3.7 Flash leads 9 (BenchLeader, 16 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 | Gemini 3.7 Flash | 62 vs 61 category score; LiveCodeBench 88.7% vs 87.3% | Medium |
| Agents & tool use | GPT-5.3-Codex | 80 vs 49 category score; AA-Briefcase v1.1 868 vs 1115 | High |
| Long documents | Either | 66 vs 66 category score; AA-LCR v1.1 81.7% vs 83.3% | Either |
| Images and other media | Gemini 3.7 Flash | 67 vs 61 category score; MMMU-Pro 85.5% vs 78.5% | Medium |
| Chat quality (human preference) | – | Only Gemini 3.7 Flash has an Arena rating (68 category score) | Not comparable |
| Lowest cost | Gemini 3.7 Flash | $1.50 vs $4.81 per 1M blended; agentic coding session $0.060 vs $0.161 | – |
| Fastest first answer | Gemini 3.7 Flash | 2.2 s (Gemini 3.7 Flash (high)) vs 38.2 s (GPT-5.3-Codex (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 3.7 Flash (high) and GPT-5.3-Codex (xhigh) are neck and neck on quality (62.1 vs 62.9, within the ±7.8 margin).
- Gemini 3.7 Flash (high) is stronger in coding, multimodal.
- GPT-5.3-Codex (xhigh) is stronger in agents & tools, long context.
- Gemini 3.7 Flash (high) is 3.2× cheaper ($1.50 vs $4.81 per 1M blended).
At a glance
| Spec | Gemini 3.7 Flash | GPT-5.3-Codex |
|---|---|---|
| Developer | OpenAI | |
| Released | 13 Aug 2026 | 5 Feb 2026 |
| Context window (tokens) | 1,048,576 | 400,000 |
| Compared at | Gemini 3.7 Flash (high) | GPT-5.3-Codex (xhigh) |
| Open weights | No | No |
| Price in / out / cached input per 1M | $0.750 / $3.75 / $0.075 | $1.75 / $14.00 / $0.175 |
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 3.7 Flash (high) | GPT-5.3-Codex (xhigh) | Leader |
|---|---|---|---|
| Reasoning | 67.8 | – | not comparable |
| Coding | 62.3 | 61.0 | Gemini 3.7 Flash |
| Agents & tools | 48.9 | 80.1 | GPT-5.3-Codex |
| Maths | 62.6 | – | not comparable |
| Knowledge | 65.9 | – | not comparable |
| Instruction following | – | 73.6 | not comparable |
| Human preference | 68.1 | – | not comparable |
| Multimodal | 67.2 | 60.8 | Gemini 3.7 Flash |
| Long context | 65.5 | 66.4 | level |
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: Reasoning, Maths, Knowledge, Instruction following, Human preference.
Where sources disagree
Across the 12 benchmarks both have, Gemini 3.7 Flash leads 9, GPT-5.3-Codex leads 3.
| Overall index | Gemini 3.7 Flash (high) | GPT-5.3-Codex (xhigh) | Leader |
|---|---|---|---|
| BenchLeader Index | 62.1 | 62.9 | 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. Cost, tokens and time per benchmark task are measured: Artificial Analysis’ averages over its whole Intelligence Index run.
| Measure | Gemini 3.7 Flash (high) | GPT-5.3-Codex (xhigh) |
|---|---|---|
| Blended price per 1M tokens | $1.50 | $4.81 |
| Input / output per 1M | $0.750 / $3.75 | $1.75 / $14.00 |
| Cached input per 1M | $0.075 | $0.175 |
| Price per task: chat reply | $0.0014 | $0.0049 |
| Price per task: summarize a 30-page report | $0.011 | $0.029 |
| Price per task: code edit | $0.010 | $0.032 |
| Price per task: agentic coding session | $0.060 | $0.161 |
| Price per task: structured extraction | $0.0023 | $0.0063 |
| Agentic coding session, input mostly cached | $0.030 | $0.090 |
| Output speed | 95 tok/s | 94 tok/s |
| Time to first answer | 2.2 s | 38.2 s |
| Time to a complete answer, last measured 13 Nov 2026 | – | 97.1 s |
Each model is compared at its best-ranked reasoning-effort setting. Gemini 3.7 Flash (high) scores 62.1 and takes 2 s to its first answer token; Gemini 3.7 Flash (medium) scores 60.1 in 2 s, Gemini 3.7 Flash (low) scores 57.3 in 2 s. GPT-5.3-Codex (xhigh) scores 62.9 and takes 38 s to its first answer token. 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
104 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 3.7 Flash (high) | GPT-5.3-Codex (xhigh) |
|---|---|---|---|
| BenchLeader Index | BenchLeader | 62.1 | 62.9 |
| Agents & tools score | BenchLeader | 48.9 | 80.1 |
| Coding score | BenchLeader | 62.3 | 61.0 |
| Human preference score | BenchLeader | 68.1 | – |
| Knowledge score | BenchLeader | 65.9 | – |
| Long context score | BenchLeader | 65.5 | 66.4 |
| Maths score | BenchLeader | 62.6 | – |
| Multimodal score | BenchLeader | 67.2 | 60.8 |
| Reasoning score | BenchLeader | 67.8 | – |
| Instruction following score | BenchLeader | – | 73.6 |
| Blended price $/M | BenchLeader | $1.50 | $4.81 |
| Output speed | BenchLeader | 95 tok/s | 94 tok/s |
| Time to first answer | BenchLeader | 2.2 s | 38.2 s |
| Context window | BenchLeader | 1,048,576 tokens | 400,000 tokens |
| Reasoning | |||
| GPQA Diamond | Epoch AI Benchmarking Hub | 94.8% | – |
| Arena Hard Prompts | Arena (LMArena) | 1510 | – |
| LiveBench 1.0 Reasoning | LiveBench | 87.8% | – |
| GPQA Diamond (AA) | Artificial Analysis | 94.5% | 91.5% |
| Humanity's Last Exam (AA) | Artificial Analysis | 47.9% | 42.5% |
| GPQA Diamond (Vals) | Vals AI | 93.9% | – |
| ARC-AGI-1 | ARC Prize | 95.5% | – |
| ARC-AGI-2 | ARC Prize | 84.6% | – |
| CritPt | Artificial Analysis | 14.3% | 16.9% |
| Arena Expert | Arena (LMArena) | 1519 | – |
| Chess Puzzles | Epoch AI Benchmarking Hub | 66.0% | – |
| Mystery Game Puzzles | Epoch AI Benchmarking Hub | 37.0% | – |
| LMCA | Epoch AI Benchmarking Hub | 50.4% | – |
| DTBench | Epoch AI Benchmarking Hub | 96.8% | – |
| ForecastBench | ForecastBench | 66.2% | – |
| Coding | |||
| SWE-bench Verified (Epoch) | Epoch AI Benchmarking Hub | – | 74.8%† |
| SciCode | SciCode | 57.2% | – |
| WeirdML | WeirdML | – | 77.9% |
| FrontierCode | Cognition | 43.6%† | – |
| Arena Coding | Arena (LMArena) | 1520 | – |
| Arena WebDev | Arena (LMArena) | 1588 | 1406† |
| LiveBench 1.0 Coding | LiveBench | 78.9% | – |
| LiveCodeBench | Vals AI | 88.7% | 87.3% |
| IOI | Vals AI | 67.8% | 53.8% |
| SWE-bench (Vals) | Vals AI | 80.8% | 78.0% |
Show all 104 benchmarks
| Metric | Publisher | Gemini 3.7 Flash (high) | GPT-5.3-Codex (xhigh) |
|---|---|---|---|
| Coding | |||
| Code Migration | Vals AI | 34.8% | – |
| ProgramBench | Vals AI | 0.0% | – |
| Vibe Code Bench v1.1 | Vals AI | 70.4% | 61.8% |
| MirrorCode | Epoch AI Benchmarking Hub | 41.1% | – |
| DeepSWE v1.1 | Epoch AI Benchmarking Hub | 65.3% | – |
| ALE-Bench | Epoch AI Benchmarking Hub | 904.3 | 1655.2 |
| FrontierSWE | Epoch AI Benchmarking Hub | 20.3% | – |
| Terminal-Bench 4.0 (AA) | Artificial Analysis | 13.6% | – |
| Terminal-Bench 2.1 (AA) | Artificial Analysis | 85.8% | – |
| Design Arena Website | Design Arena | 1295† | 1155† |
| Agents & tools | |||
| Terminal-Bench | Terminal-Bench | 12.4% | 78.4%† |
| Remote Labor Index | Scale AI / CAIS | 5.0%† | – |
| APEX-Agents | Mercor | 67.8%† | – |
| Arena Agent | Arena (LMArena) | -2.2 | – |
| LiveBench Agentic Coding | LiveBench | 58.3% | – |
| Terminal-Bench Hard | Artificial Analysis | – | 53.0% |
| τ²-Bench Telecom (AA) | Artificial Analysis | – | 86.0% |
| Terminal-Bench 2.1 (Vals) | Vals AI | 77.5% | – |
| HiL-Bench | Scale AI SEAL | – | 4.3%† |
| GDPval-AA v2.1 | Artificial Analysis | 1404 | – |
| τ³-Banking (AA) | Artificial Analysis | 32.8% | – |
| Analyst Agent (AA) | Artificial Analysis | 60.0% | – |
| CyberBench | Vals AI | 43.8% | – |
| Finance Agent v2 | Vals AI | 59.0% | – |
| Harvey's Legal Agent Benchmark v1 | Vals AI | 8.8% | – |
| Legal Research Bench | Vals AI | 34.6% | – |
| SkillsBench | Vals AI | 65.9% | – |
| SREBench | Vals AI | 4.6% | – |
| Tax Agent Bench | Vals AI | 22.8% | – |
| Terminal-Bench 2.0 (Vals) | Vals AI | – | 64.0%† |
| Terminal-Bench 4.0 (Vals) | Vals AI | 12.1% | – |
| Terminal-Bench Science | Vals AI | 7.1% | – |
| Arena Agent Task Completion | Arena (LMArena) | -3.5 | – |
| METR Time Horizons | METR | – | 74.5%† |
| Vending-Bench 2 | Andon Labs | – | 5940.1† |
| AutomationBench | Artificial Analysis | 62.0% | – |
| GDP.pdf (AA) | Artificial Analysis | 23.6% | – |
| EnterpriseOps-Gym | Artificial Analysis | 50.4%† | – |
| AA-Briefcase v1.1 | Artificial Analysis | 1115 | 868 |
| Design Arena Agentic | Design Arena | 1222† | 1018† |
| Maths | |||
| FrontierMath Tiers 1–3 | Epoch AI Benchmarking Hub | 71.6% | – |
| FrontierMath Tier 4 | Epoch AI Benchmarking Hub | 36.6% | – |
| OTIS Mock AIME | Epoch AI Benchmarking Hub | 97.2% | – |
| ProofBench | Vals AI | 58.0%† | – |
| LiveBench Mathematics | LiveBench | 93.5% | – |
| Arena Maths | Arena (LMArena) | 1502 | – |
| Knowledge | |||
| SimpleQA Verified | Epoch AI Benchmarking Hub | 69.2% | – |
| LiveBench 1.0 Data Analysis | LiveBench | 68.0% | – |
| MMLU-Pro | Vals AI | 90.1% | – |
| LegalBench | Vals AI | 87.3% | – |
| TaxEval | Vals AI | 74.7% | – |
| Excel Modeling Benchmark | Vals AI | 71.3% | – |
| MedCode | Vals AI | 53.4% | – |
| MedScribe | Vals AI | 83.9% | – |
| GDP.pdf | Epoch AI Benchmarking Hub | 23.8% | – |
| Instruction following | |||
| LiveBench Language | LiveBench | 85.5% | – |
| LiveBench Instruction Following | LiveBench | 79.9% | – |
| IFBench | Artificial Analysis | – | 75.4% |
| Arena Instruction Following | Arena (LMArena) | 1486 | – |
| Human preference | |||
| Arena Text | Arena (LMArena) | 1489 | – |
| EQ-Bench Creative Writing v3 | EQ-Bench | 1720† | – |
| Arena Creative Writing | Arena (LMArena) | 1487 | – |
| Arena Multi-turn | Arena (LMArena) | 1490 | – |
| Arena Longer Queries | Arena (LMArena) | 1501 | – |
| Multimodal | |||
| Arena Vision | Arena (LMArena) | 1313 | – |
| MMMU-Pro | Artificial Analysis | 85.5% | 78.5% |
| MMMU-Pro (Vals) | Vals AI | 89.0% | – |
| MortgageTax | Vals AI | 66.7% | – |
| Humanity's Sixth Sense | Scale AI SEAL | 39.8% | – |
| Furniture Assembly | Epoch AI Benchmarking Hub | 26.7% | – |
| Long context | |||
| AA-LCR v1.1 | Artificial Analysis | 81.7% | 83.3% |
| Arena Document | Arena (LMArena) | 1468 | – |
| MLCR | Artificial Analysis | 12.2% | – |
| Composite | |||
| Epoch Capabilities Index | Epoch AI Benchmarking Hub | 158.4† | – |
| LiveBench 1.0 | LiveBench | 78.8% | – |
| AA Intelligence Index v4.3.2 | Artificial Analysis | 39.1 | 32.5 |
| AA Progress Index v1 | Artificial Analysis | 317.4 | 301.3 |
| Vals RSI Index | Vals AI | 18.3% | – |
| Safety & honesty | |||
| DistressBench | Scale AI SEAL | 69.0%† | – |
Data as of 2026-11-16. Best configuration of each model; every score links to its source on the model pages.
What changed
- Gemini 3.7 Flash (high): 2 new results (GDPval-AA v2.1, AA-Briefcase v1.1).
- Gemini 3.7 Flash (high): BenchLeader Index 62.6 → 62.1.
- Gemini 3.7 Flash (high): 1 score revision.
- Gemini 3.7 Flash (high): 1 new result (Arena Document).
- Gemini 3.7 Flash (high): 1 new result (MirrorCode).
- GPT-5.3-Codex (xhigh): 1 new result (AA-Briefcase v1.1).
- Gemini 3.7 Flash (high): 1 new result (AA Progress Index v1).
- GPT-5.3-Codex (xhigh): 1 new result (AA Progress Index v1).
- Gemini 3.7 Flash (high): 2 new results (Harvey's Legal Agent Benchmark v1, Vals RSI Index).
- Gemini 3.7 Flash (high): input price moved from $0.38/M to $0.75/M (97% up); output price moved from $1.88/M to $3.75/M (99% up).
Related comparisons
- Gemini 3.7 Flash vs Claude Fable 5.2
- Gemini 3.7 Flash vs Gemini 4 Carbon
- Gemini 3.7 Flash vs Claude Opus 5.5
- GPT-5.3-Codex vs Claude Fable 5.2
- GPT-5.3-Codex vs Gemini 4 Carbon
- GPT-5.3-Codex vs Claude Opus 5.5
- GPT-5.3-Codex vs GPT-6 Astra
- Gemini 3.7 Flash vs Claude Fable 5.1
Rankings: Best LLM overall · Best LLM for agents · Best cheap LLMs
Gemini 3.7 Flash vs GPT-5.3-Codex: questions
- Is Gemini 3.7 Flash better than GPT-5.3-Codex?
- Gemini 3.7 Flash (high) and GPT-5.3-Codex (xhigh) are neck and neck on quality (62.1 vs 62.9, within the ±7.8 margin). The BenchLeader Index combines every independent quality benchmark; the two are within the margin of error as of 16 Nov 2026, so check the category scores for your use.
- Gemini 3.7 Flash vs GPT-5.3-Codex: which is cheaper, and will that change?
- Gemini 3.7 Flash is cheaper today: $1.50 against $4.81 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
- Is Gemini 3.7 Flash or GPT-5.3-Codex better for coding?
- Gemini 3.7 Flash scores higher in coding (62 vs 61 on the category index, where 50 is average). On LiveCodeBench, the top shared coding benchmark, it is 88.7% for Gemini 3.7 Flash against 87.3% for GPT-5.3-Codex.
- Which is faster, Gemini 3.7 Flash or GPT-5.3-Codex?
- Gemini 3.7 Flash streams faster (95 vs 94 output tokens per second, Artificial Analysis 7-day median).
- Is Gemini 3.7 Flash or GPT-5.3-Codex better for agents and tool use?
- GPT-5.3-Codex scores higher in agentic tasks (80 vs 49 on the category index, where 50 is average).
More questions (4)
- Which is better for long documents?
- They are level on long-context benchmarks (66 vs 66). Gemini 3.7 Flash accepts more context: 1,048,576 against 400,000 tokens for GPT-5.3-Codex.
- Which has lower latency?
- Gemini 3.7 Flash answers first: 2.2 s against 38.2 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 3.7 Flash vs GPT-5.3-Codex?
- Across the 12 benchmarks both have, Gemini 3.7 Flash leads 9, GPT-5.3-Codex leads 3.
- When was Gemini 3.7 Flash released?
- Gemini 3.7 Flash was released on 13 Aug 2026.