Model comparison
GPT-5.2 vs Grok 4.6
Grok 4.6 is the stronger model overall, scoring 56.9 to 54.1 on the Noometry Index.
Last verified . 40 shared benchmarks.
Summary
- They share 40 benchmarks with published results for both. GPT-5.2 scores higher in 4 categories and Grok 4.6 in 6 categories; 6 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where Grok 4.6 leads 61.4 to 50.2.
- The biggest single-benchmark swing is ProofBench: 15% for GPT-5.2 and 51% for Grok 4.6.
- Grok 4.6 is cheaper at $2 / $6 per million input/output tokens, against $1.75 / $14 for GPT-5.2.
- Grok 4.6 accepts more context: 500K tokens versus 400K.
Side by side
| GPT-5.2 | Grok 4.6 | |
|---|---|---|
| Provider | OpenAI | xAI |
| Noometry Index | 54.1 | 56.9 |
| Released | 2025-12-11 | 2026-08-12 |
| Weights | Proprietary | Proprietary |
| Context window | 400K | 500K |
| Max output | 128K | 500K |
| Input $ / M tokens | $1.75 | $2 |
| Output $ / M tokens | $14 | $6 |
| Results tracked | 67 | 49 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Grok 4.6 leads
GPT-5.2: 51.6 (#37), Grok 4.6: 58.5 (#16)
| Benchmark | GPT-5.2 | Grok 4.6 |
|---|---|---|
| LMArena WebDev | 1416 | 1617 |
| WeirdML | 72.2% | 67.3% |
| LMArena Coding | 1447 | 1465 |
| ALE-Bench | 1,294 | 1,508 |
| SWE-bench Verified | 73.8% | — |
| DeepSWE | — | 67.5% |
| FrontierCode | — | 48% |
| SWE-bench Verified (bash only) | 72.8% | — |
| CursorBench | — | 41.4% |
| SWE-bench Multilingual | 66.7% | — |
| FrontierSWE | — | 25.3% |
| SciCode | — | 56.5% |
| GSO | 27.4% | — |
| AlgoTune | 2.05 | — |
Agentic & Tool Use Too close to call
GPT-5.2: 40.2 (#24), Grok 4.6: 39.4 (#27)
| Benchmark | GPT-5.2 | Grok 4.6 |
|---|---|---|
| Vending-Bench 2 | 3,591 | 9,047 |
| Terminal-Bench | 64.9% | — |
| APEX-Agents | — | 65.3% |
| Berkeley Function Calling Leaderboard | 55.9% | — |
| GDPval | 49.7% | — |
| Remote Labor Index | 2.5% | — |
| τ²-bench Airline | 83% | — |
| τ²-bench Banking | 32.2% | — |
| τ²-bench Retail | 81.6% | — |
| τ²-bench Telecom | 89.7% | — |
| DeepResearch Bench | 41.1% | — |
| GDP.pdf | — | 17.2% |
| LMArena Search | 1207 | — |
| METR Time Horizons | 75.3% | — |
Reasoning Grok 4.6 leads
GPT-5.2: 50.2 (#35), Grok 4.6: 61.4 (#20)
| Benchmark | GPT-5.2 | Grok 4.6 |
|---|---|---|
| ARC-AGI-2 | 52.9% | 67.1% |
| SimpleBench | 45.8% | 75.9% |
| NYT Connections (extended) | 83.6% | 80% |
| ARC-AGI-1 | 86.2% | 87.5% |
| Chess Puzzles | 49% | 40% |
| EBR-Bench | 23% | 30.5% |
| LMArena Hard Prompts | 1445 | 1447 |
| Mystery Game Puzzles | 23% | 34% |
| DTBench | 90.9% | 97.3% |
| LMCA | 43.9% | 48.5% |
| Epoch Capabilities Index | 153.45 | 156.44 |
| Kagi LLM Benchmark | 73.3% | — |
| CritPt | — | 19.7% |
| EnigmaEval | 10.4% | — |
| ForecastBench | 60.1 | — |
Math Grok 4.6 leads
GPT-5.2: 60.0 (#38), Grok 4.6: 67.0 (#24)
| Benchmark | GPT-5.2 | Grok 4.6 |
|---|---|---|
| FrontierMath (Tiers 1-3) | 67.4% | 66% |
| FrontierMath Tier 4 | 31.7% | 31.7% |
| OTIS Mock AIME 2024-2025 | 96.1% | 99.2% |
| ProofBench | 15% | 51% |
| LMArena Math | 1440 | 1423 |
| MathArena Final-Answer Competitions | 72% | — |
| FrontierMath (Feb 2025 set) | 40.7% | — |
| FrontierMath Tier 4 (v1) | 18.8% | — |
Knowledge Grok 4.6 leads
GPT-5.2: 59.3 (#32), Grok 4.6: 63.3 (#20)
| Benchmark | GPT-5.2 | Grok 4.6 |
|---|---|---|
| GPQA Diamond | 91.4% | 94% |
| SimpleQA Verified | 37.1% | 49.3% |
| LMArena Expert | 1445 | 1467 |
| Humanity's Last Exam | 27.8% | — |
| Vectara Hallucination Rate | 8.4% | — |
Multimodal GPT-5.2 leads
GPT-5.2: 51.3 (#7), Grok 4.6: 43.6 (#23)
| Benchmark | GPT-5.2 | Grok 4.6 |
|---|---|---|
| LMArena Vision | 1268 | 1263 |
| Furniture Assembly | 38.3% | 40% |
| LMArena Document | 1405 | 1452 |
| VPCT | 84% | — |
| Blueprint-Bench 2 | — | 33.2% |
Multilingual Too close to call
GPT-5.2: 53.4 (#67), Grok 4.6: 53.0 (#74)
| Benchmark | GPT-5.2 | Grok 4.6 |
|---|---|---|
| LMArena Non-English | 1425 | 1420 |
| LMArena Chinese | 1460 | 1480 |
| LMArena French | 1455 | 1461 |
| LMArena German | 1448 | 1431 |
| LMArena Japanese | 1420 | 1376 |
| LMArena Korean | 1392 | 1397 |
| LMArena Russian | 1440 | 1422 |
| LMArena Spanish | 1433 | 1404 |
Instruction Following Too close to call
GPT-5.2: 74.7 (#89), Grok 4.6: 75.4 (#63)
| Benchmark | GPT-5.2 | Grok 4.6 |
|---|---|---|
| LMArena Instruction Following | 1417 | 1431 |
Long Context Too close to call
GPT-5.2: 44.0 (#78), Grok 4.6: 44.5 (#66)
| Benchmark | GPT-5.2 | Grok 4.6 |
|---|---|---|
| LMArena Longer Query | 1428 | 1454 |
| CL-bench | 18.2% | — |
Writing & Preference GPT-5.2 leads
GPT-5.2: 66.8 (#32), Grok 4.6: 62.3 (#80)
| Benchmark | GPT-5.2 | Grok 4.6 |
|---|---|---|
| LMArena Text | 1439 | 1428 |
| LMArena Creative Writing | 1401 | 1428 |
| LMArena Multi-Turn | 1458 | 1425 |
| EQ-Bench Creative Writing | 1703 | — |
Frequently asked questions
Is GPT-5.2 better than Grok 4.6?
Grok 4.6 is the stronger model overall, scoring 56.9 to 54.1 on the Noometry Index.
Which is cheaper, GPT-5.2 or Grok 4.6?
Grok 4.6 is cheaper. It lists at $2 per million input tokens and $6 per million output tokens; GPT-5.2 lists at $1.75 and $14.
Is GPT-5.2 or Grok 4.6 better for coding?
Grok 4.6 scores higher on coding benchmarks: 58.5 versus 51.6 in the Noometry coding category.
Which has the bigger context window?
Grok 4.6 does, with 500K tokens against 400K.
How many benchmarks do GPT-5.2 and Grok 4.6 share?
40 benchmarks have published results for both models. GPT-5.2 has 67 scored results on Noometry and Grok 4.6 has 49.