GPT-5.2-Codex vs GPT-6 Sol
Full profiles: GPT-5.2-Codex (OpenAI) and GPT-6 Sol (OpenAI): every benchmark, price and rank.
Updated · Data from 14 sources
As of 16 Nov 2026, GPT-5.2-Codex and GPT-6 Sol are neck and neck on quality (57.6 vs 62.6 on the BenchLeader Index, inside the ±10.5 margin), and GPT-6 Sol is 1.2× cheaper at $4.00 vs $4.81 per 1M tokens blended. Choose GPT-5.2-Codex for agents & tools (69 vs 59) and fast first answers (1.5 s vs 36.9 s); choose GPT-6 Sol for coding (66 vs 55), where it also streams ~2× faster.
Key facts (as of 16 Nov 2026)
- GPT-5.2-Codex scores 57.6 on the BenchLeader Index and GPT-6 Sol scores 62.6 (BenchLeader, 16 Nov 2026).
- GPT-5.2-Codex costs $4.81 per 1M tokens blended (list price, 16 Nov 2026).
- GPT-6 Sol costs $4.00 per 1M tokens blended (list price, 16 Nov 2026).
- GPT-6 Sol streams ~2× faster than GPT-5.2-Codex (Artificial Analysis, 7-day median, 16 Nov 2026).
- Of the 11 benchmarks both models have, GPT-6 Sol leads 9 (BenchLeader, 16 Nov 2026).
- On Arena WebDev, GPT-5.2-Codex is rated 1339 and GPT-6 Sol 1673 (LMArena, as published by 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 | GPT-6 Sol | 66 vs 55 category score; Arena WebDev 1673 vs 1339 | High |
| Agents & tool use | GPT-5.2-Codex | 69 vs 59 category score; Terminal-Bench 66.5% vs 50.3% | High |
| Chat quality (human preference) | – | Only GPT-6 Sol has an Arena rating (64 category score) | Not comparable |
| Lowest cost | GPT-6 Sol | $4.00 vs $4.81 per 1M blended; agentic coding session $0.160 vs $0.161 | – |
| Fastest first answer | GPT-5.2-Codex | 1.5 s (GPT-5.2-Codex) vs 36.9 s (GPT-6 Sol (max)) 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
- GPT-5.2-Codex and GPT-6 Sol (max) are neck and neck on quality (57.6 vs 62.6, within the ±10.5 margin).
- GPT-5.2-Codex is stronger in agents & tools.
- GPT-6 Sol (max) is stronger in coding.
- GPT-6 Sol (max) is 1.2× cheaper ($4.00 vs $4.81 per 1M blended).
- GPT-6 Sol (max) streams 2.8× faster (127 vs 45 tokens per second).
At a glance
| Spec | GPT-5.2-Codex | GPT-6 Sol |
|---|---|---|
| Developer | OpenAI | OpenAI |
| Released | 18 Dec 2025 | 22 Sep 2026 |
| Context window (tokens) | 272,000 | 1,050,000 |
| Compared at | GPT-5.2-Codex | GPT-6 Sol (max) |
| Open weights | No | No |
| Price in / out / cached input per 1M | $1.75 / $14.00 / $0.175 | $2.00 / $10.00 / $0.200 |
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 | GPT-5.2-Codex | GPT-6 Sol (max) | Leader |
|---|---|---|---|
| Reasoning | – | 66.0 | not comparable |
| Coding | 55.2 | 66.3 | GPT-6 Sol |
| Agents & tools | 69.0 | 58.8 | GPT-5.2-Codex |
| Maths | – | 68.0 | not comparable |
| Knowledge | – | 60.6 | not comparable |
| Human preference | – | 64.2 | not comparable |
| Multimodal | – | 60.6 | not comparable |
| Long context | – | 66.6 | 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.
Dashed: that model has no result on the axis it crosses; the line only joins its neighbours.
Not drawn, no result for these models: Instruction following.
Where sources disagree
Across the 11 benchmarks both have, GPT-5.2-Codex leads 2, GPT-6 Sol leads 9.
| Overall index | GPT-5.2-Codex | GPT-6 Sol (max) | Leader |
|---|---|---|---|
| BenchLeader Index | 57.6 | 62.6 | 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 | GPT-5.2-Codex | GPT-6 Sol (max) |
|---|---|---|
| Blended price per 1M tokens | $4.81 | $4.00 |
| Input / output per 1M | $1.75 / $14.00 | $2.00 / $10.00 |
| Cached input per 1M | $0.175 | $0.200 |
| Price per task: chat reply | $0.0049 | $0.0038 |
| Price per task: summarize a 30-page report | $0.029 | $0.030 |
| Price per task: code edit | $0.032 | $0.027 |
| Price per task: agentic coding session | $0.161 | $0.160 |
| Price per task: structured extraction | $0.0063 | $0.0060 |
| Agentic coding session, input mostly cached | $0.090 | $0.079 |
| Cost per benchmark task (AA Index) | – | $2.47 |
| Output tokens per task (AA Index) | – | 54k |
| Of which reasoning | – | 35k |
| Time per task (AA Index) | – | 7 min |
| Output speed | 45 tok/s | 127 tok/s |
| Time to first answer | 1.5 s | 36.9 s |
| Time to a complete answer, last measured 13 Nov 2026 | – | 143.8 s |
Each model is compared at its best-ranked reasoning-effort setting. GPT-5.2-Codex scores 57.6 and takes 1 s to its first answer token. GPT-6 Sol (max) scores 62.6 and takes 37 s to its first answer token; GPT-6 Sol (xhigh) scores 58.4 in 19 s, GPT-6 Sol (high) scores 58.1 in 6 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
126 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 | GPT-5.2-Codex | GPT-6 Sol (max) |
|---|---|---|---|
| BenchLeader Index | BenchLeader | 57.6 | 62.6 |
| Agents & tools score | BenchLeader | 69.0 | 58.8 |
| Coding score | BenchLeader | 55.2 | 66.3 |
| Human preference score | BenchLeader | – | 64.2 |
| Knowledge score | BenchLeader | – | 60.6 |
| Long context score | BenchLeader | – | 66.6 |
| Maths score | BenchLeader | – | 68.0 |
| Multimodal score | BenchLeader | – | 60.6 |
| Reasoning score | BenchLeader | – | 66.0 |
| Blended price $/M | BenchLeader | $4.81 | $4.00 |
| Output speed | BenchLeader | 45 tok/s | 127 tok/s |
| Time to first answer | BenchLeader | 1.5 s | 36.9 s |
| Context window | BenchLeader | 272,000 tokens | 1,050,000 tokens |
| Reasoning | |||
| GPQA Diamond | Epoch AI Benchmarking Hub | – | 94.3% |
| SimpleBench | SimpleBench | – | 73.1%† |
| Arena Hard Prompts | Arena (LMArena) | – | 1483 |
| LiveBench 1.0 Reasoning | LiveBench | 77.7% | 88.7% |
| GPQA Diamond (AA) | Artificial Analysis | 89.9%† | – |
| Humanity's Last Exam (AA) | Artificial Analysis | 35.7%† | 47.9% |
| HLE-Diamond (AA) | Artificial Analysis | – | 45.2% |
| ARC-AGI-1 | ARC Prize | – | 95.5% |
| ARC-AGI-2 | ARC Prize | – | 89.6% |
| ARC-AGI-3 | ARC Prize | – | 23.0% |
| CritPt | Artificial Analysis | 8.7%† | 30.9% |
| CritPt v2 hard subset (AA) | Artificial Analysis | – | 66.5% |
| MysteryMechanism | Vals AI | – | 30.2% |
| HLE-Diamond (Scale) | Scale AI SEAL | – | 33.8%† |
| Arena Expert | Arena (LMArena) | – | 1496 |
| Chess Puzzles | Epoch AI Benchmarking Hub | – | 56.0% |
| Mystery Game Puzzles | Epoch AI Benchmarking Hub | – | 56.0% |
| LMCA | Epoch AI Benchmarking Hub | – | 59.1% |
| DTBench | Epoch AI Benchmarking Hub | – | 97.3% |
| Coding | |||
| SciCode | SciCode | – | 57.6% |
| FrontierCode | Cognition | – | 49.3% |
| Arena Coding | Arena (LMArena) | – | 1521 |
| Arena WebDev | Arena (LMArena) | 1339 | 1673 |
| LiveBench 1.0 Coding | LiveBench | 83.6% | 81.8% |
| LiveCodeBench | Vals AI | 88.0% | – |
Show all 126 benchmarks
| Metric | Publisher | GPT-5.2-Codex | GPT-6 Sol (max) |
|---|---|---|---|
| Coding | |||
| IOI | Vals AI | – | 82.6% |
| SWE-bench (Vals) | Vals AI | 72.4%† | – |
| SWE-Bench Pro | Scale AI SEAL | 41.0% | – |
| SWE-bench Verified (bash only) | SWE-bench | 72.8% | – |
| SWE-bench Multilingual (bash only) | SWE-bench | 66.3% | – |
| SWE-bench Verified (any scaffold) | SWE-bench | 72.8% | – |
| Code Migration | Vals AI | – | 57.2% |
| ProgramBench | Vals AI | – | 2.0% |
| Vibe Code Bench v1.1 | Vals AI | 37.9%† | 87.8% |
| SWE-Bench Pro (private) | Scale AI SEAL | 27.7% | – |
| MirrorCode | Epoch AI Benchmarking Hub | – | 48.9% |
| ALE-Bench | Epoch AI Benchmarking Hub | 1299.9 | 2462 |
| WeirdML v3 | WeirdML | – | 19.7%† |
| Terminal-Bench 4.0 (AA) | Artificial Analysis | – | 43.9% |
| GameBench 2 Overall | DuelLab GameBench 2 | – | 73.0† |
| GameBench 2 by reasoning setting | DuelLab GameBench 2 | – | 79.4† |
| Design Arena Website | Design Arena | – | 1290† |
| FrontierCode 2.0 (AA) | Artificial Analysis | – | 35.6% |
| Agents & tools | |||
| Terminal-Bench | Terminal-Bench | 66.5% | 50.3% |
| APEX-Agents | Mercor | – | 54.3% |
| Arena Agent | Arena (LMArena) | – | 6.9 |
| LiveBench Agentic Coding | LiveBench | 49.4% | 52.9% |
| Terminal-Bench Hard | Artificial Analysis | 37.1%† | – |
| τ²-Bench Telecom (AA) | Artificial Analysis | 92.1%† | – |
| Terminal-Bench 2.1 (Vals) | Vals AI | – | 83.2% |
| GDPval-AA v2.1 | Artificial Analysis | – | 1514 |
| ITBench SRE (AA) | Artificial Analysis | – | 49.4% |
| CyberBench | Vals AI | – | 78.0% |
| Finance Agent v2 | Vals AI | – | 49.0% |
| Harvey's Legal Agent Benchmark v1 | Vals AI | – | 1.7% |
| Legal Research Bench | Vals AI | – | 28.9% |
| SREBench | Vals AI | – | 30.5% |
| Tax Agent Bench | Vals AI | – | 15.0% |
| Terminal-Bench 4.0 (Vals) | Vals AI | – | 44.4% |
| Terminal-Bench Science | Vals AI | – | 30.0% |
| Arena Search | Arena (LMArena) | – | 1229 |
| Arena Agent Task Completion | Arena (LMArena) | – | 4.5 |
| Vending-Bench 2 | Andon Labs | – | 14427.9† |
| SpireClimb | Epoch AI Benchmarking Hub | – | 9.2% |
| AutomationBench | Artificial Analysis | – | 61.6% |
| GDP.pdf (AA) | Artificial Analysis | – | 25.2% |
| Harvey LAB-AA v1.1 | Harvey | – | 3.6% |
| AA-Briefcase v1.1 | Artificial Analysis | – | 1476 |
| Design Arena Agentic | Design Arena | 1039 | 1218† |
| Terminal-Bench-Science 0.1 (AA) | Artificial Analysis | – | 30.0% |
| CWE-Bench (AA) | Artificial Analysis | – | 64.2% |
| DeepSec-Bench (AA) | Artificial Analysis | – | 46.3% |
| CyberGym end-to-end (AA) | Artificial Analysis | – | 0.0% |
| FACTS Search | – | 92.7%† | |
| Maths | |||
| FrontierMath Tiers 1–3 | Epoch AI Benchmarking Hub | – | 89.8% |
| FrontierMath Tier 4 | Epoch AI Benchmarking Hub | – | 90.0% |
| OTIS Mock AIME | Epoch AI Benchmarking Hub | – | 100.0% |
| ProofBench | Vals AI | – | 83.0%† |
| LiveBench Mathematics | LiveBench | 88.8% | 96.4% |
| Arena Maths | Arena (LMArena) | – | 1440 |
| FrontierMath Erdős | Epoch AI Benchmarking Hub | – | 1.5% |
| Knowledge | |||
| SimpleQA Verified | Epoch AI Benchmarking Hub | – | 60.7% |
| LiveBench 1.0 Data Analysis | LiveBench | 78.2% | 81.2% |
| AA-Omniscience v1.1 | Artificial Analysis | – | 21.4 |
| PRBench Legal | Scale AI SEAL | – | 38.6%† |
| BioMysteryBench | Vals AI | – | 74.8% |
| Excel Modeling Benchmark | Vals AI | – | 71.5% |
| MedCode | Vals AI | – | 47.1% |
| MedScribe | Vals AI | – | 82.0% |
| Public Benefits Bench v1.1 | Vals AI | – | 56.6% |
| Financial Coaching Bench | Vals AI | – | 62.6% |
| EBR-bench | Epoch AI Benchmarking Hub | – | 53.3% |
| GDP.pdf | Epoch AI Benchmarking Hub | – | 26.4% |
| AA-Omniscience v1.1: accuracy | Artificial Analysis | – | 51.1% |
| AA-Omniscience v1.1: non-hallucination | Artificial Analysis | – | 39.1% |
| HealthBench Professional (AA) | OpenAI | – | 62.3% |
| FACTS Parametric | – | 76.1%† | |
| Instruction following | |||
| LiveBench Language | LiveBench | 73.7% | 85.3% |
| LiveBench Instruction Following | LiveBench | 66.5% | 68.6% |
| IFBench | Artificial Analysis | 77.6%† | – |
| Arena Instruction Following | Arena (LMArena) | – | 1449 |
| Human preference | |||
| Arena Text | Arena (LMArena) | – | 1457 |
| EQ-Bench Creative Writing v3 | EQ-Bench | – | 2126† |
| Arena Creative Writing | Arena (LMArena) | – | 1428 |
| Arena Multi-turn | Arena (LMArena) | – | 1468 |
| Arena Longer Queries | Arena (LMArena) | – | 1466 |
| Multimodal | |||
| Arena Vision | Arena (LMArena) | – | 1233 |
| MMMU-Pro | Artificial Analysis | 76.3%† | 83.0% |
| Humanity's Sixth Sense | Scale AI SEAL | – | 31.2% |
| Blueprint-Bench 2 | Epoch AI Benchmarking Hub | – | 36.9%† |
| Furniture Assembly | Epoch AI Benchmarking Hub | – | 58.3% |
| AA-3D | Artificial Analysis | – | 1131 |
| FACTS Multimodal | – | 43.9%† | |
| Long context | |||
| AA-LCR v1.1 | Artificial Analysis | 82.3%† | 83.7% |
| MLCR | Artificial Analysis | – | 33.9% |
| FACTS Grounding | – | 83.3%† | |
| Composite | |||
| Epoch Capabilities Index | Epoch AI Benchmarking Hub | – | 163.8† |
| LiveBench 1.0 | LiveBench | 74.0% | 79.3% |
| AA Intelligence Index v5.0.1 | Artificial Analysis | – | 44.2 |
| AA Intelligence Index v4.3.2 | Artificial Analysis | 28.5† | 47.6 |
| AA Progress Index v1 | Artificial Analysis | 290.2† | 334 |
| Vals RSI Index | Vals AI | – | 28.1% |
| Vals Science Index | Vals AI | – | 45.0% |
| AA Cyber Index | Artificial Analysis | – | 36.8 |
| Safety & honesty | |||
| FORTRESS | Scale AI SEAL | – | 13.5%† |
| DistressBench | Scale AI SEAL | – | 68.0%† |
Data as of 2026-11-16. Best configuration of each model; every score links to its source on the model pages.
DuelLab GameBench 2, release gb2-20261111-7b598bb13de4 (https://benchmarks.duellab.org/releases/gb2-20261111-7b598bb13de4/), CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Scores mapped to BenchLeader model names; not endorsed by DuelLab.
What changed
- GPT-6 Sol (max): 1 new result (GDPval-AA v2.1).
- GPT-6 Sol (max): 1 new result (MirrorCode).
- GPT-6 Sol (max): BenchLeader Index 63.5 → 62.9.
- GPT-6 Sol (max): 2 score revisions.
- GPT-6 Sol (max): 1 score revision.
- GPT-6 Sol (max): 1 score revision.
- GPT-6 Sol (max): 6 new results (AA Intelligence Index v5.0.1, AA-Omniscience v1.1, CritPt v2 hard subset (AA)…).
- GPT-6 Sol (max): 1 score revision.
- GPT-6 Sol (max): 1 score revision.
- GPT-6 Sol (max): BenchLeader Index 64.1 → 63.5.
Related comparisons
- GPT-5.2-Codex vs Claude Opus 5.5
- GPT-5.2-Codex vs GPT-6.1 Sol
- GPT-5.2-Codex vs Claude Fable 5
- GPT-6 Sol vs Claude Fable 5.2
- GPT-6 Sol vs Gemini 4 Carbon
- GPT-6 Sol vs Claude Opus 5.5
- GPT-6 Sol vs GPT-6 Astra
- GPT-5.2-Codex vs Gemini 4 Argon
Rankings: Best LLM overall · Best LLM for agents · Best cheap LLMs
GPT-5.2-Codex vs GPT-6 Sol: questions
- Is GPT-5.2-Codex better than GPT-6 Sol?
- GPT-5.2-Codex and GPT-6 Sol (max) are neck and neck on quality (57.6 vs 62.6, within the ±10.5 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.
- GPT-5.2-Codex vs GPT-6 Sol: which is cheaper, and will that change?
- GPT-6 Sol is cheaper today: $4.00 against $4.81 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
- Is GPT-5.2-Codex or GPT-6 Sol better for coding?
- GPT-6 Sol scores higher in coding (66 vs 55 on the category index, where 50 is average). On Arena WebDev, the top shared coding benchmark, it is 1339 for GPT-5.2-Codex against 1673 for GPT-6 Sol.
- Which is faster, GPT-5.2-Codex or GPT-6 Sol?
- GPT-6 Sol streams ~2× faster (127 vs 45 output tokens per second, Artificial Analysis 7-day median).
- Is GPT-5.2-Codex or GPT-6 Sol better for agents and tool use?
- GPT-5.2-Codex scores higher in agentic tasks (69 vs 59 on the category index, where 50 is average). On Terminal-Bench, the top shared agents & tools benchmark, it is 66.5% for GPT-5.2-Codex against 50.3% for GPT-6 Sol.
More questions (4)
- Which is better for long documents?
- GPT-6 Sol accepts more context: 1,050,000 against 272,000 tokens for GPT-5.2-Codex.
- Which has lower latency?
- GPT-5.2-Codex answers first: 1.5 s against 36.9 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 GPT-5.2-Codex vs GPT-6 Sol?
- Across the 11 benchmarks both have, GPT-5.2-Codex leads 2, GPT-6 Sol leads 9.
- When was GPT-6 Sol released?
- GPT-6 Sol was released on 22 Sep 2026.