GPT-5.6 Sol vs Grok 4.20
Full profiles: GPT-5.6 Sol (OpenAI) and Grok 4.20 (SpaceXAI): every benchmark, price and rank.
Updated · Data from 12 sources
As of 6 Nov 2026, GPT-5.6 Sol is ahead of Grok 4.20 on quality (65.4 vs 58.4 on the BenchLeader Index), but Grok 4.20 is 5.1× cheaper at $1.56 vs $8.00 per 1M tokens blended. Choose GPT-5.6 Sol for maths (71 vs 54), agents & tools (67 vs 51), and coding (66 vs 54); choose Grok 4.20 for fast first answers (19.9 s vs 69.6 s) and instruction following (79 vs 71).
Key facts (as of 6 Nov 2026)
- GPT-5.6 Sol scores 65.4 on the BenchLeader Index and Grok 4.20 scores 58.4 (BenchLeader, 6 Nov 2026).
- GPT-5.6 Sol costs $8.00 per 1M tokens blended (list price, 6 Nov 2026).
- Grok 4.20 costs $1.56 per 1M tokens blended (list price, 6 Nov 2026).
- Of the 40 benchmarks both models have, GPT-5.6 Sol leads 37 (BenchLeader, 6 Nov 2026).
GPT-5.6 Sol benchmarks 
Grok 4.20 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 | GPT-5.6 Sol | 66 vs 54 category score; LiveCodeBench 82.6% vs 84.3% | High |
| Agents & tool use | GPT-5.6 Sol | 67 vs 51 category score; Terminal-Bench Hard 65.9% vs 37.9% | High |
| Maths | GPT-5.6 Sol | 71 vs 54 category score; FrontierMath Tiers 1–3 89.1% vs 44.9% | High |
| Reasoning | GPT-5.6 Sol | 66 vs 61 category score; GPQA Diamond 93.5% vs 89.3% | Medium |
| Knowledge | GPT-5.6 Sol | 64 vs 53 category score; SimpleQA Verified 69.7% vs 30.2% | High |
| Long documents | GPT-5.6 Sol | 67 vs 59 category score; AA-LCR v1.1 84.0% vs 69.0% | High |
| Images and other media | GPT-5.6 Sol | 66 vs 58 category score; MMMU-Pro 83.4% vs 74.6% | High |
| Following instructions | Grok 4.20 | 79 vs 71 category score; IFBench 81.2% vs 72.7% | High |
| Chat quality (human preference) | – | Only Grok 4.20 has an Arena rating (66 category score) | Not comparable |
| Lowest cost | Grok 4.20 | $1.56 vs $8.00 per 1M blended; agentic coding session $0.085 vs $0.320 | – |
| Fastest first answer | Grok 4.20 | 19.9 s (Grok 4.20 (thinking)) vs 69.6 s (GPT-5.6 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-5.6 Sol (max) leads on quality: 65.4 vs 58.4.
- GPT-5.6 Sol (max) is stronger in agents & tools, coding, knowledge, long context, maths, multimodal, reasoning.
- Grok 4.20 (thinking) is stronger in instruction following, human preference.
- Grok 4.20 (thinking) is 5.1× cheaper ($1.56 vs $8.00 per 1M blended).
At a glance
| Spec | GPT-5.6 Sol | Grok 4.20 |
|---|---|---|
| Developer | OpenAI | SpaceXAI |
| Released | 9 Jul 2026 | 17 Feb 2026 |
| Context window (tokens) | 1,050,000 | 2,000,000 |
| Compared at | GPT-5.6 Sol (max) | Grok 4.20 (thinking) |
| Open weights | No | No |
| Price in / out / cached input per 1M | $4.00 / $20.00 / $0.400 | $1.25 / $2.50 / $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-5.6 Sol (max) | Grok 4.20 (thinking) | Leader |
|---|---|---|---|
| Reasoning | 65.6 | 61.0 | GPT-5.6 Sol |
| Coding | 65.8 | 53.5 | GPT-5.6 Sol |
| Agents & tools | 66.5 | 50.5 | GPT-5.6 Sol |
| Maths | 70.7 | 53.9 | GPT-5.6 Sol |
| Knowledge | 64.2 | 53.0 | GPT-5.6 Sol |
| Instruction following | 71.1 | 78.6 | Grok 4.20 |
| Human preference | – | 66.0 | not comparable |
| Multimodal | 66.0 | 58.3 | GPT-5.6 Sol |
| Long context | 66.9 | 59.1 | GPT-5.6 Sol |
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: Human preference.
ModelIndex
Where sources disagree
Across the 40 benchmarks both have, GPT-5.6 Sol leads 37, Grok 4.20 leads 3. Every overall index that rates both puts GPT-5.6 Sol ahead.
| Overall index | GPT-5.6 Sol (max) | Grok 4.20 (thinking) | Leader |
|---|---|---|---|
| BenchLeader Index | 65.4 | 58.4 | GPT-5.6 Sol |
| Arena Text | 1485† | 1472 | GPT-5.6 Sol |
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-5.6 Sol (max) | Grok 4.20 (thinking) |
|---|---|---|
| Blended price per 1M tokens | $8.00 | $1.56 |
| Input / output per 1M | $4.00 / $20.00 | $1.25 / $2.50 |
| Cached input per 1M | $0.400 | $0.200 |
| Price per task: chat reply | $0.0076 | $0.0013 |
| Price per task: summarize a 30-page report | $0.060 | $0.017 |
| Price per task: code edit | $0.054 | $0.011 |
| Price per task: agentic coding session | $0.320 | $0.085 |
| Price per task: structured extraction | $0.012 | $0.0030 |
| Agentic coding session, input mostly cached | $0.158 | $0.038 |
| Output speed | 101 tok/s | 102 tok/s |
| Time to first answer | 69.6 s | 19.9 s |
| Time to a complete answer | 169.6 s | 50.3 s |
Each model is compared at its best-ranked reasoning-effort setting. GPT-5.6 Sol (max) scores 65.4 and takes 70 s to its first answer token; GPT-5.6 Sol (xhigh) scores 64.2 in 28 s, GPT-5.6 Sol (high) scores 63.6 in 14 s. Grok 4.20 (thinking) scores 58.4 and takes 20 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
138 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-5.6 Sol (max) | Grok 4.20 (thinking) |
|---|---|---|---|
| BenchLeader Index | BenchLeader | 65.4 | 58.4 |
| Agents & tools score | BenchLeader | 66.5 | 50.5 |
| Coding score | BenchLeader | 65.8 | 53.5 |
| Instruction following score | BenchLeader | 71.1 | 78.6 |
| Knowledge score | BenchLeader | 64.2 | 53.0 |
| Long context score | BenchLeader | 66.9 | 59.1 |
| Maths score | BenchLeader | 70.7 | 53.9 |
| Multimodal score | BenchLeader | 66.0 | 58.3 |
| Reasoning score | BenchLeader | 65.6 | 61.0 |
| Human preference score | BenchLeader | – | 66.0 |
| Blended price $/M | BenchLeader | $8.00 | $1.56 |
| Output speed | BenchLeader | 101 tok/s | 102 tok/s |
| Time to first answer | BenchLeader | 69.6 s | 19.9 s |
| Context window | BenchLeader | 1,050,000 tokens | 2,000,000 tokens |
| Reasoning | |||
| GPQA Diamond | Epoch AI Benchmarking Hub | 93.5% | 89.3% |
| SimpleBench | SimpleBench | 64.8%† | – |
| Arena Hard Prompts | Arena (LMArena) | 1511† | 1489 |
| LiveBench 1.0 Reasoning | LiveBench | 91.7% | – |
| GPQA Diamond (AA) | Artificial Analysis | 94.1% | 91.1% |
| Humanity's Last Exam (AA) | Artificial Analysis | 49.5% | 34.5% |
| GPQA Diamond (Vals) | Vals AI | 95.2% | 88.6% |
| EnigmaEval | Scale AI SEAL | 37.1%† | – |
| Kagi LLM Benchmark | Kagi LLM Benchmark | 67.0%† | 75.0% |
| ARC-AGI-1 | ARC Prize | 96.5% | 89.5% |
| ARC-AGI-2 | ARC Prize | 92.5% | 65.1% |
| ARC-AGI-3 | ARC Prize | 7.8% | 0.1% |
| CritPt | Artificial Analysis | 32.3% | 6.6% |
| CritPt v2 hard subset (AA) | Artificial Analysis | 60.0% | – |
| MysteryMechanism | Vals AI | 33.3% | – |
| Chess Puzzles | Epoch AI Benchmarking Hub | 59.0% | 29.0% |
| Mystery Game Puzzles | Epoch AI Benchmarking Hub | 58.0% | – |
| CL-bench | Epoch AI Benchmarking Hub | – | 22.2%† |
| CL-bench Life | Epoch AI Benchmarking Hub | – | 11.9%† |
| LMCA | Epoch AI Benchmarking Hub | 58.4% | 38.7%† |
| DTBench | Epoch AI Benchmarking Hub | 95.5% | 90.1%† |
| ForecastBench | ForecastBench | 65.5% | 64.5% |
| BTF-3 | Epoch AI Benchmarking Hub | 13.7% | – |
| Coding | |||
| SciCode | SciCode | 57.1% | – |
| WeirdML | WeirdML | 87.0% | 52.3%† |
Show all 138 benchmarks
| Metric | Publisher | GPT-5.6 Sol (max) | Grok 4.20 (thinking) |
|---|---|---|---|
| Coding | |||
| FrontierCode | Cognition | 47.5%† | – |
| GSO-Bench | GSO-Bench | 76.5%† | – |
| Arena Coding | Arena (LMArena) | 1533† | 1511 |
| Arena WebDev | Arena (LMArena) | 1616† | 1376 |
| LiveBench 1.0 Coding | LiveBench | 83.9% | – |
| LiveCodeBench | Vals AI | 82.6% | 84.3% |
| IOI | Vals AI | 91.2% | – |
| SWE-bench (Vals) | Vals AI | 96.2% | 72.2% |
| Code Migration | Vals AI | 52.9% | 0.3% |
| ProgramBench | Vals AI | 1.5% | – |
| Vibe Code Bench 1-100 | Vals AI | 20.0% | – |
| Vibe Code Bench v1.1 | Vals AI | 80.5% | 4.1% |
| SWE Atlas: Codebase QnA | Scale AI SEAL | 46.0%† | – |
| SWE Atlas: Test Writing | Scale AI SEAL | 45.9%† | – |
| MirrorCode | Epoch AI Benchmarking Hub | 20.0%† | – |
| Surface Evolver Bench | Epoch AI Benchmarking Hub | 93.1%† | – |
| DeepSWE v1.1 | Epoch AI Benchmarking Hub | 72.7% | – |
| CursorBench | Cursor | 41.7% | – |
| ALE-Bench | Epoch AI Benchmarking Hub | 2176.9 | 1150.3† |
| FrontierSWE | Epoch AI Benchmarking Hub | 32.2% | – |
| Terminal-Bench 4.0 (AA) | Artificial Analysis | 39.9% | – |
| Terminal-Bench 2.1 (AA) | Artificial Analysis | 88.0% | – |
| Startrise Frontend | Startrise | 66.2%† | – |
| GameBench 2 Overall | DuelLab GameBench 2 | 66.7† | – |
| GameBench 2 by reasoning setting | DuelLab GameBench 2 | 79.6† | – |
| Agents & tools | |||
| Terminal-Bench | Terminal-Bench | 37.3% | 57.3%† |
| OSWorld-Verified 2.0 | OSWorld | 27.3% | – |
| APEX-Agents | Mercor | 51.4%† | – |
| Arena Agent | Arena (LMArena) | 4.6† | – |
| LiveBench Agentic Coding | LiveBench | 56.2% | – |
| Terminal-Bench Hard | Artificial Analysis | 65.9% | 37.9% |
| τ²-Bench Telecom (AA) | Artificial Analysis | 85.1% | 93.0% |
| Terminal-Bench 2.1 (Vals) | Vals AI | 85.8% | 44.2% |
| HiL-Bench | Scale AI SEAL | 32.3%† | 20.0%† |
| τ³-bench Banking | τ²-bench | 46.9%† | – |
| τ³-Banking (AA) | Artificial Analysis | 44.3% | – |
| ITBench SRE (AA) | Artificial Analysis | 56.2% | – |
| Analyst Agent (AA) | Artificial Analysis | 47.5% | – |
| APEX-Agents (AA) | Artificial Analysis | – | 14.2% |
| CUA-bench | Vals AI | 8.3% | – |
| CyberBench | Vals AI | 76.3% | – |
| Finance Agent v2 | Vals AI | 53.8% | 28.5% |
| Harvey's Legal Agent Benchmark v1 | Vals AI | 2.5% | 0.0% |
| Legal Research Bench | Vals AI | 48.1% | 13.9% |
| SkillsBench | Vals AI | 54.1% | – |
| SREBench | Vals AI | 30.5% | – |
| Tax Agent Bench | Vals AI | 31.1% | 7.0% |
| Terminal-Bench 2.0 (Vals) | Vals AI | – | 40.5% |
| Terminal-Bench 4.0 (Vals) | Vals AI | 37.9% | – |
| Terminal-Bench Science | Vals AI | 20.0% | – |
| Time Horizon Index: KSP | Vals AI | 23.8% | – |
| Web Search Index | Vals AI | 43.6% | – |
| Arena Search | Arena (LMArena) | 1258† | – |
| BALROG | BALROG | 60.0% | – |
| PostTrainBench | Epoch AI Benchmarking Hub | 36.2% | – |
| GBAEval | Epoch AI Benchmarking Hub | 52.6%† | – |
| Vending-Bench 2 | Andon Labs | 9619.4† | 4662.9† |
| AutomationBench | Artificial Analysis | 60.1% | – |
| GDP.pdf (AA) | Artificial Analysis | 27.2% | – |
| EnterpriseOps-Gym | Artificial Analysis | 42.9% | – |
| AA-Briefcase v1.1 | Artificial Analysis | 1479 | – |
| Terminal-Bench-Science 0.1 (AA) | Artificial Analysis | 22.4% | – |
| Maths | |||
| FrontierMath Tiers 1–3 | Epoch AI Benchmarking Hub | 89.1% | 44.9% |
| FrontierMath Tier 4 | Epoch AI Benchmarking Hub | 82.9% | 17.1% |
| OTIS Mock AIME | Epoch AI Benchmarking Hub | 100.0% | 92.2% |
| ProofBench | Vals AI | 83.0% | 14.0% |
| LiveBench Mathematics | LiveBench | 96.2% | – |
| AIME (Vals) | Vals AI | – | 96.5% |
| Arena Maths | Arena (LMArena) | 1495† | 1465 |
| Knowledge | |||
| SimpleQA Verified | Epoch AI Benchmarking Hub | 69.7% | 30.2% |
| LiveBench 1.0 Data Analysis | LiveBench | 79.8% | – |
| MMLU-Pro | Vals AI | 89.1% | 86.3% |
| LegalBench | Vals AI | 87.0% | 77.7% |
| CorpFin | Vals AI | 64.4% | 63.7% |
| TaxEval | Vals AI | 74.8% | 74.1% |
| MedQA | Vals AI | – | 94.5% |
| PRBench Finance | Scale AI SEAL | 50.5% | – |
| PRBench Legal | Scale AI SEAL | 50.5% | – |
| BioMysteryBench | Vals AI | 71.1% | – |
| CaseLaw v2 | Vals AI | – | 54.5% |
| Excel Modeling Benchmark | Vals AI | 72.3% | 11.7% |
| MedCode | Vals AI | 44.0% | 32.2% |
| MedScribe | Vals AI | 85.2% | 63.4% |
| Public Benefits Bench v1.1 | Vals AI | 66.5% | – |
| DrugDiscoveryBench | Scale AI SEAL | 56.5% | – |
| EBR-bench | Epoch AI Benchmarking Hub | 44.8% | – |
| GDP.pdf | Epoch AI Benchmarking Hub | 30.7% | – |
| Global-MMLU-Lite (AA) | Artificial Analysis | – | 89.5% |
| Instruction following | |||
| LiveBench Language | LiveBench | 87.7% | – |
| LiveBench Instruction Following | LiveBench | 71.8% | – |
| IFBench | Artificial Analysis | 72.7% | 81.2% |
| Arena Instruction Following | Arena (LMArena) | 1491† | 1447 |
| Human preference | |||
| Arena Text | Arena (LMArena) | 1485† | 1472 |
| EQ-Bench 4 | EQ-Bench | 1236† | – |
| Arena Creative Writing | Arena (LMArena) | 1470† | 1445 |
| Arena Multi-turn | Arena (LMArena) | 1490† | 1479 |
| Arena Longer Queries | Arena (LMArena) | 1498† | 1466 |
| Multimodal | |||
| Arena Vision | Arena (LMArena) | 1281† | 1262 |
| MMMU-Pro | Artificial Analysis | 83.4% | 74.6% |
| MMMU-Pro (Vals) | Vals AI | 88.8% | 83.5% |
| MortgageTax | Vals AI | 67.3% | 45.4% |
| Blueprint-Bench 2 | Epoch AI Benchmarking Hub | 33.6%† | 0.0%† |
| Long context | |||
| AA-LCR v1.1 | Artificial Analysis | 84.0% | 69.0% |
| Arena Document | Arena (LMArena) | 1489† | 1439 |
| MLCR | Artificial Analysis | 30.0% | – |
| Composite | |||
| Epoch Capabilities Index | Epoch AI Benchmarking Hub | 163.0† | 152.3† |
| LiveBench 1.0 | LiveBench | 81.0% | – |
| AA Intelligence Index v4.3.2 | Artificial Analysis | 47.0 | 25.7 |
| AA Progress Index v1 | Artificial Analysis | 331.5 | 281.9 |
| Vals RSI Index | Vals AI | 23.9% | – |
| Vals Science Index | Vals AI | 41.5% | – |
| Vals Multimodal Index | Vals AI | 72.6% | 39.1% |
| Safety & honesty | |||
| FORTRESS | Scale AI SEAL | 20.1%† | – |
Data as of 2026-11-06. Best configuration of each model; every score links to its source on the model pages.
DuelLab GameBench 2, release gb2-20261103-09b2218cbd85 (https://benchmarks.duellab.org/releases/gb2-20261103-09b2218cbd85/), CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Scores mapped to BenchLeader model names; not endorsed by DuelLab.
What changed
- GPT-5.6 Sol (max): 1 score revision.
- Grok 4.20 (thinking): BenchLeader Index 58.9 → 58.4.
- GPT-5.6 Sol (max): 2 new results (AA Progress Index v1, CritPt v2 hard subset (AA)).
- Grok 4.20 (thinking): 1 new result (AA Progress Index v1).
- GPT-5.6 Sol (max): 1 new result (AA-Briefcase v1.1).
- Grok 4.20 (thinking): 1 new result (Global-MMLU-Lite (AA)).
- GPT-5.6 Sol (max): 3 new results (Harvey's Legal Agent Benchmark v1, Vals RSI Index, Vals Science Index).
- Grok 4.20 (thinking): 1 new result (Harvey's Legal Agent Benchmark v1).
- GPT-5.6 Sol (max): 1 score revision.
- GPT-5.6 Sol (max): 1 new result (ForecastBench).
Related comparisons
- GPT-5.6 Sol vs Claude Opus 5.5
- GPT-5.6 Sol vs Gemini 4 Carbon
- GPT-5.6 Sol vs GPT-6 Astra
- Grok 4.20 vs Claude Opus 5.5
- Grok 4.20 vs Gemini 4 Carbon
- Grok 4.20 vs GPT-6 Astra
- Grok 4.20 vs GPT-6.1 Sol
- GPT-5.6 Sol vs Claude Fable 5.1
Rankings: Best LLM overall · Best LLM for agents · Best cheap LLMs
GPT-5.6 Sol vs Grok 4.20: questions
- Is GPT-5.6 Sol better than Grok 4.20?
- GPT-5.6 Sol (max) leads on quality: 65.4 vs 58.4. The BenchLeader Index combines every independent quality benchmark; GPT-5.6 Sol (max) is ahead overall as of 6 Nov 2026, but check the category scores for your use.
- GPT-5.6 Sol vs Grok 4.20: which is cheaper, and will that change?
- Grok 4.20 is cheaper today: $1.56 against $8.00 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
- Is GPT-5.6 Sol or Grok 4.20 better for coding?
- GPT-5.6 Sol scores higher in coding (66 vs 54 on the category index, where 50 is average). On LiveCodeBench, the top shared coding benchmark, it is 82.6% for GPT-5.6 Sol against 84.3% for Grok 4.20.
- Which is faster, GPT-5.6 Sol or Grok 4.20?
- Grok 4.20 streams faster (102 vs 101 output tokens per second, Artificial Analysis 7-day median).
- Is GPT-5.6 Sol or Grok 4.20 better for agents and tool use?
- GPT-5.6 Sol scores higher in agentic tasks (67 vs 51 on the category index, where 50 is average). On Terminal-Bench Hard, the top shared agents & tools benchmark, it is 65.9% for GPT-5.6 Sol against 37.9% for Grok 4.20.
More questions (4)
- Which is better for long documents?
- GPT-5.6 Sol scores higher on long-context benchmarks (67 vs 59). Grok 4.20 accepts more context: 2,000,000 against 1,050,000 tokens for GPT-5.6 Sol.
- Which has lower latency?
- Grok 4.20 answers first: 19.9 s against 69.6 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-5.6 Sol vs Grok 4.20?
- Across the 40 benchmarks both have, GPT-5.6 Sol leads 37, Grok 4.20 leads 3. Every overall index that rates both puts GPT-5.6 Sol ahead.
- When was GPT-5.6 Sol released?
- GPT-5.6 Sol was released on 9 Jul 2026.