Model comparison
Grok 4.6 vs Qwen3.8 27B
Grok 4.6 is the stronger model overall, scoring 56.9 to 46.0 on the Noometry Index. Qwen3.8 27B costs 2.7× less per token, which makes it the better buy when Grok 4.6's lead doesn't matter for your workload.
Last verified . 29 shared benchmarks.
Alibaba (Qwen)Summary
- They share 29 benchmarks with published results for both. Grok 4.6 scores higher in 7 categories and Qwen3.8 27B in 3 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in math, where Grok 4.6 leads 67.0 to 37.1.
- The biggest single-benchmark swing is ProofBench: 51% for Grok 4.6 and 16% for Qwen3.8 27B.
- Qwen3.8 27B is cheaper at $0.99 / $1.49 per million input/output tokens, against $2 / $6 for Grok 4.6.
- Grok 4.6 accepts more context: 500K tokens versus 262K.
- Qwen3.8 27B has downloadable open weights; the other is API-only.
Side by side
| Grok 4.6 | Qwen3.8 27B | |
|---|---|---|
| Provider | xAI | Alibaba (Qwen) |
| Noometry Index | 56.9 | 46.0 |
| Released | 2026-08-12 | 2026-08-14 |
| Weights | Proprietary | Open |
| Context window | 500K | 262K |
| Max output | 500K | 33K |
| Input $ / M tokens | $2 | $0.99 |
| Output $ / M tokens | $6 | $1.49 |
| Results tracked | 49 | 31 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Grok 4.6 leads
Grok 4.6: 58.5 (#16), Qwen3.8 27B: 50.5 (#44)
| Benchmark | Grok 4.6 | Qwen3.8 27B |
|---|---|---|
| LMArena WebDev | 1617 | 1593 |
| SciCode | 56.5% | 46.6% |
| LMArena Coding | 1465 | 1482 |
| DeepSWE | 67.5% | — |
| FrontierCode | 48% | — |
| CursorBench | 41.4% | — |
| FrontierSWE | 25.3% | — |
| WeirdML | 67.3% | — |
| ALE-Bench | 1,508 | — |
Agentic & Tool Use Grok 4.6 leads
Grok 4.6: 39.4 (#27), Qwen3.8 27B: 32.9 (#57)
| Benchmark | Grok 4.6 | Qwen3.8 27B |
|---|---|---|
| APEX-Agents | 65.3% | 47.5% |
| GDP.pdf | 17.2% | — |
| Vending-Bench 2 | 9,047 | — |
Reasoning Grok 4.6 leads
Grok 4.6: 61.4 (#20), Qwen3.8 27B: 41.0 (#54)
| Benchmark | Grok 4.6 | Qwen3.8 27B |
|---|---|---|
| ARC-AGI-2 | 67.1% | 42.4% |
| NYT Connections (extended) | 80% | 54.5% |
| ARC-AGI-1 | 87.5% | 87.5% |
| CritPt | 19.7% | 5.4% |
| LMArena Hard Prompts | 1447 | 1460 |
| DTBench | 97.3% | 88% |
| LMCA | 48.5% | 41.4% |
| Epoch Capabilities Index | 156.44 | 149.38 |
| SimpleBench | 75.9% | — |
| Chess Puzzles | 40% | — |
| EBR-Bench | 30.5% | — |
| Mystery Game Puzzles | 34% | — |
| Surface Evolver Bench | — | 45% |
Math Grok 4.6 leads
Grok 4.6: 67.0 (#24), Qwen3.8 27B: 37.1 (#161)
| Benchmark | Grok 4.6 | Qwen3.8 27B |
|---|---|---|
| ProofBench | 51% | 16% |
| LMArena Math | 1423 | 1456 |
| FrontierMath (Tiers 1-3) | 66% | — |
| FrontierMath Tier 4 | 31.7% | — |
| OTIS Mock AIME 2024-2025 | 99.2% | — |
Knowledge Grok 4.6 leads
Grok 4.6: 63.3 (#20), Qwen3.8 27B: 41.6 (#109)
| Benchmark | Grok 4.6 | Qwen3.8 27B |
|---|---|---|
| LMArena Expert | 1467 | 1482 |
| GPQA Diamond | 94% | — |
| SimpleQA Verified | 49.3% | — |
Multimodal Grok 4.6 leads
Grok 4.6: 43.6 (#23), Qwen3.8 27B: 41.3 (#37)
| Benchmark | Grok 4.6 | Qwen3.8 27B |
|---|---|---|
| LMArena Vision | 1263 | 1271 |
| Blueprint-Bench 2 | 33.2% | — |
| Furniture Assembly | 40% | — |
| LMArena Document | 1452 | — |
Multilingual Too close to call
Grok 4.6: 53.0 (#74), Qwen3.8 27B: 53.7 (#60)
| Benchmark | Grok 4.6 | Qwen3.8 27B |
|---|---|---|
| LMArena Non-English | 1420 | 1430 |
| LMArena Chinese | 1480 | 1504 |
| LMArena French | 1461 | 1465 |
| LMArena German | 1431 | 1438 |
| LMArena Japanese | 1376 | 1384 |
| LMArena Korean | 1397 | 1393 |
| LMArena Russian | 1422 | 1415 |
| LMArena Spanish | 1404 | 1448 |
Instruction Following Too close to call
Grok 4.6: 75.4 (#63), Qwen3.8 27B: 75.8 (#53)
| Benchmark | Grok 4.6 | Qwen3.8 27B |
|---|---|---|
| LMArena Instruction Following | 1431 | 1439 |
Long Context Too close to call
Grok 4.6: 44.5 (#66), Qwen3.8 27B: 44.3 (#70)
| Benchmark | Grok 4.6 | Qwen3.8 27B |
|---|---|---|
| LMArena Longer Query | 1454 | 1450 |
Writing & Preference Qwen3.8 27B leads
Grok 4.6: 62.3 (#80), Qwen3.8 27B: 65.8 (#43)
| Benchmark | Grok 4.6 | Qwen3.8 27B |
|---|---|---|
| LMArena Text | 1428 | 1441 |
| LMArena Creative Writing | 1428 | 1384 |
| LMArena Multi-Turn | 1425 | 1441 |
| EQ-Bench Creative Writing | — | 1671 |
Frequently asked questions
Is Grok 4.6 better than Qwen3.8 27B?
Grok 4.6 is the stronger model overall, scoring 56.9 to 46.0 on the Noometry Index. Qwen3.8 27B costs 2.7× less per token, which makes it the better buy when Grok 4.6's lead doesn't matter for your workload.
Which is cheaper, Grok 4.6 or Qwen3.8 27B?
Qwen3.8 27B is cheaper. It lists at $0.99 per million input tokens and $1.49 per million output tokens; Grok 4.6 lists at $2 and $6.
Is Grok 4.6 or Qwen3.8 27B better for coding?
Grok 4.6 scores higher on coding benchmarks: 58.5 versus 50.5 in the Noometry coding category.
Which has the bigger context window?
Grok 4.6 does, with 500K tokens against 262K.
How many benchmarks do Grok 4.6 and Qwen3.8 27B share?
29 benchmarks have published results for both models. Grok 4.6 has 49 scored results on Noometry and Qwen3.8 27B has 31.