Model comparison
gpt-oss-20b vs Grok 4.6
Grok 4.6 is the stronger model overall, scoring 56.9 to 32.5 on the Noometry Index. gpt-oss-20b costs 83× less per token, which makes it the better buy when Grok 4.6's lead doesn't matter for your workload.
Last verified . 26 shared benchmarks.
Summary
- They share 26 benchmarks with published results for both. gpt-oss-20b scores higher in 0 categories and Grok 4.6 in 9 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where Grok 4.6 leads 61.4 to 19.3.
- The biggest single-benchmark swing is Chess Puzzles: 4% for gpt-oss-20b and 40% for Grok 4.6.
- gpt-oss-20b is cheaper at $0.018 / $0.09 per million input/output tokens, against $2 / $6 for Grok 4.6.
- Grok 4.6 accepts more context: 500K tokens versus 131K.
- gpt-oss-20b has downloadable open weights; the other is API-only.
Side by side
| gpt-oss-20b | Grok 4.6 | |
|---|---|---|
| Provider | OpenAI | xAI |
| Noometry Index | 32.5 | 56.9 |
| Released | 2025-08-05 | 2026-08-12 |
| Weights | Open | Proprietary |
| Context window | 131K | 500K |
| Max output | 16K | 500K |
| Input $ / M tokens | $0.018 | $2 |
| Output $ / M tokens | $0.09 | $6 |
| Results tracked | 34 | 49 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Grok 4.6 leads
gpt-oss-20b: 37.6 (#192), Grok 4.6: 58.5 (#16)
| Benchmark | gpt-oss-20b | Grok 4.6 |
|---|---|---|
| SciCode | 34.4% | 56.5% |
| WeirdML | 40.9% | 67.3% |
| LMArena Coding | 1306 | 1465 |
| ALE-Bench | 566.05 | 1,508 |
| DeepSWE | — | 67.5% |
| FrontierCode | — | 48% |
| CursorBench | — | 41.4% |
| LMArena WebDev | — | 1617 |
| FrontierSWE | — | 25.3% |
Agentic & Tool Use Grok 4.6 leads
gpt-oss-20b: 9.3 (#154), Grok 4.6: 39.4 (#27)
| Benchmark | gpt-oss-20b | Grok 4.6 |
|---|---|---|
| Terminal-Bench | 3.4% | — |
| APEX-Agents | — | 65.3% |
| GDP.pdf | — | 17.2% |
| Vending-Bench 2 | — | 9,047 |
Reasoning Grok 4.6 leads
gpt-oss-20b: 19.3 (#261), Grok 4.6: 61.4 (#20)
| Benchmark | gpt-oss-20b | Grok 4.6 |
|---|---|---|
| CritPt | 1.4% | 19.7% |
| Chess Puzzles | 4% | 40% |
| LMArena Hard Prompts | 1274 | 1447 |
| DTBench | 68% | 97.3% |
| LMCA | 14.5% | 48.5% |
| Epoch Capabilities Index | 137.82 | 156.44 |
| ARC-AGI-2 | — | 67.1% |
| SimpleBench | — | 75.9% |
| Kagi LLM Benchmark | 53.2% | — |
| NYT Connections (extended) | — | 80% |
| ARC-AGI-1 | — | 87.5% |
| EBR-Bench | — | 30.5% |
| Mystery Game Puzzles | — | 34% |
Math Grok 4.6 leads
gpt-oss-20b: 39.4 (#103), Grok 4.6: 67.0 (#24)
| Benchmark | gpt-oss-20b | Grok 4.6 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 65.3% | 99.2% |
| LMArena Math | 1317 | 1423 |
| FrontierMath (Tiers 1-3) | — | 66% |
| FrontierMath Tier 4 | — | 31.7% |
| ProofBench | — | 51% |
| Omni-MATH | 56.5% | — |
Knowledge Grok 4.6 leads
gpt-oss-20b: 34.6 (#195), Grok 4.6: 63.3 (#20)
| Benchmark | gpt-oss-20b | Grok 4.6 |
|---|---|---|
| GPQA Diamond | 60.8% | 94% |
| LMArena Expert | 1258 | 1467 |
| SimpleQA Verified | — | 49.3% |
| MMLU-Pro | 74% | — |
| GPQA (HELM) | 59.4% | — |
Multimodal Not comparable
gpt-oss-20b: —, Grok 4.6: 43.6 (#23)
| Benchmark | gpt-oss-20b | Grok 4.6 |
|---|---|---|
| LMArena Vision | — | 1263 |
| Blueprint-Bench 2 | — | 33.2% |
| Furniture Assembly | — | 40% |
| LMArena Document | — | 1452 |
Multilingual Grok 4.6 leads
gpt-oss-20b: 42.2 (#197), Grok 4.6: 53.0 (#74)
| Benchmark | gpt-oss-20b | Grok 4.6 |
|---|---|---|
| LMArena Non-English | 1268 | 1420 |
| LMArena Chinese | 1314 | 1480 |
| LMArena German | 1255 | 1431 |
| LMArena Japanese | 1244 | 1376 |
| LMArena Korean | 1236 | 1397 |
| LMArena Russian | 1278 | 1422 |
| LMArena Spanish | 1267 | 1404 |
| LMArena French | — | 1461 |
Instruction Following Grok 4.6 leads
gpt-oss-20b: 61.8 (#240), Grok 4.6: 75.4 (#63)
| Benchmark | gpt-oss-20b | Grok 4.6 |
|---|---|---|
| LMArena Instruction Following | 1236 | 1431 |
| IFEval | 73.2% | — |
Long Context Grok 4.6 leads
gpt-oss-20b: 37.9 (#209), Grok 4.6: 44.5 (#66)
| Benchmark | gpt-oss-20b | Grok 4.6 |
|---|---|---|
| LMArena Longer Query | 1250 | 1454 |
Writing & Preference Grok 4.6 leads
gpt-oss-20b: 35.5 (#265), Grok 4.6: 62.3 (#80)
| Benchmark | gpt-oss-20b | Grok 4.6 |
|---|---|---|
| LMArena Text | 1287 | 1428 |
| LMArena Creative Writing | 1201 | 1428 |
| LMArena Multi-Turn | 1268 | 1425 |
| EQ-Bench Creative Writing | 666 | — |
| WildBench | 73.7% | — |
Frequently asked questions
Is gpt-oss-20b better than Grok 4.6?
Grok 4.6 is the stronger model overall, scoring 56.9 to 32.5 on the Noometry Index. gpt-oss-20b costs 83× less per token, which makes it the better buy when Grok 4.6's lead doesn't matter for your workload.
Which is cheaper, gpt-oss-20b or Grok 4.6?
gpt-oss-20b is cheaper. It lists at $0.018 per million input tokens and $0.09 per million output tokens; Grok 4.6 lists at $2 and $6.
Is gpt-oss-20b or Grok 4.6 better for coding?
Grok 4.6 scores higher on coding benchmarks: 58.5 versus 37.6 in the Noometry coding category.
Which has the bigger context window?
Grok 4.6 does, with 500K tokens against 131K.
How many benchmarks do gpt-oss-20b and Grok 4.6 share?
26 benchmarks have published results for both models. gpt-oss-20b has 34 scored results on Noometry and Grok 4.6 has 49.