Grok 4.5 vs Qwen3 235B A22B Thinking 2507
Grok 4.5 leads Qwen3 235B A22B Thinking 2507 on the Artificial Analysis Intelligence Index, 38.8 to 12.7, and Qwen3 235B A22B Thinking 2507 costs 1.7x less per output token ($3.50 vs $6.00 per million). Updated October 22, 2026; 43 benchmarks have a result for both.
Grok 4.5 or Qwen3 235B A22B Thinking 2507: which to pick
Only Grok 4.5 has a coding score, 58.6 on the Vibe Coding Index; Qwen3 235B A22B Thinking 2507 has none, so the two cannot be ranked on coding.
Pick Qwen3 235B A22B Thinking 2507 for output tokens 1.7x cheaper ($3.50 vs $6.00 per million) and work measured by LMArena Text: medicine and healthcare, where it leads.
- Shared results
- 43
- Grok 4.5 only
- 81
- Qwen3 235B A22B Thinking 2507 only
- 3
- Like-for-like categories
- 8 of 9
Only benchmarks that tested both models count toward a pick; a result only one model has is shown but not compared.
Which one for your work
Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.
Coding
Vibe Coding Index, six public coding benchmarks
No call
Only Grok 4.5 has a coding score, so no call is made.
One side not measured
Intelligence
AA Intelligence Index
Grok 4.5
Grok 4.5 leads, 38.8 against 12.7.
Same board, both measured
Agentic work
AA Agentic Index
Grok 4.5
Grok 4.5 leads, 41.2 against 1.3.
Same board, both measured
Reasoning
GPQA Diamond
No call
Only Grok 4.5 has a GPQA Diamond result, 92.9%.
Directional only
Knowledge
MMLU Pro
No call
Only Grok 4.5 has a MMLU Pro result, 89.2%.
Directional only
Web design
Design Arena: websites
Grok 4.5
Grok 4.5 leads, 1281 against 1052.
Same board, both measured
Long documents
Listed context window
Grok 4.5
Grok 4.5 takes 500K tokens against 128K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Qwen3 235B A22B Thinking 2507
Grok 4.5 $0.0050, Qwen3 235B A22B Thinking 2507 $0.0022 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only Grok 4.5 has the number, $0.62.
Directional only
At a glance
- Coding index
- Grok 4.5 58.6 from 6 benchmarks, Qwen3 235B A22B Thinking 2507 not benchmarked.
- Widest category
- Web design, 35.2 points apart on average over 6 shared benchmarks.
- Widest single gap
- AA Coding Index: Grok 4.5 72.4 against Qwen3 235B A22B Thinking 2507 22.1, 50.3 points apart on a 0 to 100 scale.
- Output price
- Grok 4.5 $6.00 against Qwen3 235B A22B Thinking 2507 $3.50 per million output tokens, 1.7x apart.
- Context window
- Grok 4.5 takes 500K tokens, Qwen3 235B A22B Thinking 2507 128K.
Benchmark by benchmark
First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.
| Category | Grok 4.5 | Bars | Qwen3 235B A22B Thinking 2507 | Ties |
|---|---|---|---|---|
| Coding | 4 | 0 | ||
| Intelligence | 1 | 0 | ||
| Agentic work | 2 | 0 | ||
| Reasoning | 4 | 0 | ||
| Math | 3 | 0 | ||
| Knowledge | 4 | 1 | ||
| Web design | 6 | 0 | ||
| General | 18 | 0 | ||
| All shared boards | 42 | 1 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index40
| Benchmark | Grok 4.5 | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| LiveBench coding | 94.3% | Not measured | Only Grok 4.5 measured |
| LMArena WebDev | 1553 | Not measured | Only Grok 4.5 measured |
| AA Coding Index | 72.4 | 22.1 | Grok 4.5 |
| LiveCodeBench (Vals AI) | 87.3% | Not measured | Only Grok 4.5 measured |
| SWE-bench Verified (Vals AI) | 86.6% | Not measured | Only Grok 4.5 measured |
| Terminal-Bench 4 (Vals AI) | 8.6% | Not measured | Only Grok 4.5 measured |
| Vibe Code Bench (Vals AI) | 69.0% | Not measured | Only Grok 4.5 measured |
| Code migration (Vals AI) | 36.6% | Not measured | Only Grok 4.5 measured |
| ProgramBench (Vals AI) | 0.0% | Not measured | Only Grok 4.5 measured |
| IOI (Vals AI) | 40.6% | Not measured | Only Grok 4.5 measured |
| Terminal-Bench 2.1 (Vals AI) | 67.8% | Not measured | Only Grok 4.5 measured |
| SkillsBench (Vals AI) | 66.0% | Not measured | Only Grok 4.5 measured |
| CyberBench (Vals AI) | 71.0% | Not measured | Only Grok 4.5 measured |
| SciCode | 55.0% | Not measured | Only Grok 4.5 measured |
| FrontierCode | 42.4% | Not measured | Only Grok 4.5 measured |
| WeirdML | 46.4% | 41.0% | Grok 4.5 |
| ALE-Bench | 1309.3 | Not measured | Only Grok 4.5 measured |
| LMArena Text: coding | 1474 | 1424 | Grok 4.5 |
| LMArena Text: software and IT | 1471 | 1431 | Grok 4.5 |
| LMArena WebDev: React | 1556 | Not measured | Only Grok 4.5 measured |
| LMArena WebDev: HTML | 1567 | Not measured | Only Grok 4.5 measured |
| LMArena WebDev: from an image | 1562 | Not measured | Only Grok 4.5 measured |
| SciCode | 6.5% | Not measured | Only Grok 4.5 measured |
| Neither model has a result on the other 20 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | Grok 4.5 | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| AA Intelligence Index | 38.8 | 12.7 | Grok 4.5 |
| Epoch Capabilities Index | 154.1 | Not measured | Only Grok 4.5 measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index20
| Benchmark | Grok 4.5 | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| AA Agentic Index | 41.2 | 1.3 | Grok 4.5 |
| Design Arena: agents building web apps | 1218 | Not measured | Only Grok 4.5 measured |
| Design Arena: agents building full-stack apps | 1249 | Not measured | Only Grok 4.5 measured |
| Design Arena: agents building mobile apps | 1220 | Not measured | Only Grok 4.5 measured |
| Design Arena: agents building Android apps | 1236 | Not measured | Only Grok 4.5 measured |
| Design Arena: agents building Godot games | 1229 | Not measured | Only Grok 4.5 measured |
| Design Arena: agents building games | 1236 | Not measured | Only Grok 4.5 measured |
| Design Arena: HTML slides | 1155 | Not measured | Only Grok 4.5 measured |
| Design Arena: PowerPoint slides | 1209 | Not measured | Only Grok 4.5 measured |
| LiveBench research | 85.9% | Not measured | Only Grok 4.5 measured |
| LiveBench browser | 44.0% | Not measured | Only Grok 4.5 measured |
| LiveBench long agentic | 27.9% | Not measured | Only Grok 4.5 measured |
| SREBench (Vals AI) | 0.8% | Not measured | Only Grok 4.5 measured |
| Finance Agent (Vals AI) | 48.4% | Not measured | Only Grok 4.5 measured |
| Tax Agent Bench (Vals AI) | 24.9% | Not measured | Only Grok 4.5 measured |
| Legal Research Bench (Vals AI) | 38.0% | Not measured | Only Grok 4.5 measured |
| Excel Modeling Benchmark (Vals AI) | 52.9% | Not measured | Only Grok 4.5 measured |
| Time Horizon Index (Vals AI) | 6.3% | Not measured | Only Grok 4.5 measured |
| APEX-Agents | 56.2% | Not measured | Only Grok 4.5 measured |
| PostTrainBench | 23.4% | Not measured | Only Grok 4.5 measured |
| Vending-Bench 2 | $3,887 | -$11 | Grok 4.5 |
| LMArena Search | 1216 | Not measured | Only Grok 4.5 measured |
| Neither model has a result on the other 20 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index40
| Benchmark | Grok 4.5 | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| GPQA Diamond | 92.9% | Not measured | Only Grok 4.5 measured |
| LiveBench reasoning | 87.2% | Not measured | Only Grok 4.5 measured |
| MMMU Pro (Vals AI) | 61.8% | Not measured | Only Grok 4.5 measured |
| GPQA Diamond (Artificial Analysis run) | 93.1% | Not measured | Only Grok 4.5 measured |
| Humanity's Last Exam (Artificial Analysis run) | 42.7% | Not measured | Only Grok 4.5 measured |
| GPQA Diamond (Epoch AI run) | 93.4% | 80.1% | Grok 4.5 |
| ARC-AGI | 87.2% | Not measured | Only Grok 4.5 measured |
| ARC-AGI-2 | 52.6% | Not measured | Only Grok 4.5 measured |
| SimpleBench | 70.0% | Not measured | Only Grok 4.5 measured |
| Chess puzzles | 35.0% | Not measured | Only Grok 4.5 measured |
| Mystery game puzzles | Not measured | 9.0% | Only Qwen3 235B A22B Thinking 2507 measured |
| LMArena Text: hard prompts | 1462 | 1416 | Grok 4.5 |
| LMArena Text: expert | 1465 | 1463 | Grok 4.5 |
| LMArena Text: hard prompts in English | 1464 | 1432 | Grok 4.5 |
| LMArena Vision: diagrams | 1304 | Not measured | Only Grok 4.5 measured |
| LMArena Vision: homework | 1295 | Not measured | Only Grok 4.5 measured |
| Kagi LLM Benchmark | 83.5% | Not measured | Only Grok 4.5 measured |
| Neither model has a result on the other 6 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index30
| Benchmark | Grok 4.5 | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| OTIS Mock AIME | 97.8% | 86.7% | Grok 4.5 |
| FrontierMath, tiers 1 to 3 | 57.2% | Not measured | Only Grok 4.5 measured |
| FrontierMath, tier 4 | 24.4% | Not measured | Only Grok 4.5 measured |
| ProofBench | 31.0% | Not measured | Only Grok 4.5 measured |
| LMArena Text: math | 1460 | 1412 | Grok 4.5 |
| LMArena Text: mathematical work | 1482 | 1435 | Grok 4.5 |
| MathArena | 34.1% | Not measured | Only Grok 4.5 measured |
| Neither model has a result on the other 7 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index41
| Benchmark | Grok 4.5 | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| MMLU Pro | 89.2% | Not measured | Only Grok 4.5 measured |
| LegalBench (Vals AI) | 86.0% | Not measured | Only Grok 4.5 measured |
| CorpFin (Vals AI) | 67.4% | Not measured | Only Grok 4.5 measured |
| TaxEval (Vals AI) | 71.7% | Not measured | Only Grok 4.5 measured |
| MortgageTax (Vals AI) | 61.8% | Not measured | Only Grok 4.5 measured |
| MedCode (Vals AI) | 43.3% | Not measured | Only Grok 4.5 measured |
| SimpleQA Verified | 48.3% | 40.4% | Grok 4.5 |
| LMArena Text: science | 1464 | 1433 | Grok 4.5 |
| LMArena Text: medicine and healthcare | 1440 | 1456 | Qwen3 235B A22B Thinking 2507 |
| LMArena Text: legal and government | 1463 | 1415 | Grok 4.5 |
| LMArena Text: business and finance | 1441 | 1413 | Grok 4.5 |
| FACTS Benchmark Suite | 50.2% | Not measured | Only Grok 4.5 measured |
| Neither model has a result on the other 8 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index60
| Benchmark | Grok 4.5 | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| Design Arena: websites | 1281 | 1052 | Grok 4.5 |
| Design Arena: UI components | 1277 | 942 | Grok 4.5 |
| Design Arena: game dev | 1263 | 956 | Grok 4.5 |
| Design Arena: data viz | 1266 | 950 | Grok 4.5 |
| Design Arena: 3D | 1257 | 997 | Grok 4.5 |
| Design Arena: all code categories | 1275 | 1032 | Grok 4.5 |
| Design Arena: SVG | 1214 | Not measured | Only Grok 4.5 measured |
| Design Arena: ASCII art | 1257 | Not measured | Only Grok 4.5 measured |
| LiveBench design | 78.1% | Not measured | Only Grok 4.5 measured |
GeneralLike-for-likePublic boards, shown as published, kept out of the index180
| Benchmark | Grok 4.5 | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| LiveBench data analysis | 73.0% | Not measured | Only Grok 4.5 measured |
| SAGE (Vals AI) | 35.0% | Not measured | Only Grok 4.5 measured |
| MedScribe (Vals AI) | 86.9% | Not measured | Only Grok 4.5 measured |
| Vals Multimodal Index | 63.4% | Not measured | Only Grok 4.5 measured |
| Long context reasoning (AA-LCR) | 79.3% | Not measured | Only Grok 4.5 measured |
| DTBench | 96.5% | 80.3% | Grok 4.5 |
| LMCA | 45.2% | 29.3% | Grok 4.5 |
| Fiction.LiveBench | Not measured | 75.0% | Only Qwen3 235B A22B Thinking 2507 measured |
| Creative writing (Lech Mazur) | Not measured | 82.4% | Only Qwen3 235B A22B Thinking 2507 measured |
| Surface Evolver Bench | 74.4% | Not measured | Only Grok 4.5 measured |
| Furniture assembly | 22.5% | Not measured | Only Grok 4.5 measured |
| LMArena Text | 1448 | 1415 | Grok 4.5 |
| LMArena Text: instruction following | 1445 | 1386 | Grok 4.5 |
| LMArena Text: creative writing | 1442 | 1384 | Grok 4.5 |
| LMArena Vision | 1287 | Not measured | Only Grok 4.5 measured |
| LMArena Text: English | 1453 | 1429 | Grok 4.5 |
| LMArena Text: non-English | 1440 | 1397 | Grok 4.5 |
| LMArena Text: Chinese | 1495 | 1481 | Grok 4.5 |
| LMArena Text: German | 1448 | 1408 | Grok 4.5 |
| LMArena Text: Japanese | 1430 | 1386 | Grok 4.5 |
| LMArena Text: Korean | 1406 | 1375 | Grok 4.5 |
| LMArena Text: French | 1457 | Not measured | Only Grok 4.5 measured |
| LMArena Text: Spanish | 1451 | 1389 | Grok 4.5 |
| LMArena Text: Russian | 1449 | 1401 | Grok 4.5 |
| LMArena Text: Polish | 1432 | 1390 | Grok 4.5 |
| LMArena Text: multi-turn | 1456 | 1406 | Grok 4.5 |
| LMArena Text: longer queries | 1462 | 1400 | Grok 4.5 |
| LMArena Text: writing and language | 1445 | 1388 | Grok 4.5 |
| LMArena Text: entertainment and media | 1434 | 1370 | Grok 4.5 |
| LMArena Vision: English | 1285 | Not measured | Only Grok 4.5 measured |
| LMArena Vision: Chinese | 1345 | Not measured | Only Grok 4.5 measured |
| LMArena Vision: reading text | 1293 | Not measured | Only Grok 4.5 measured |
| LMArena Vision: humor | 1276 | Not measured | Only Grok 4.5 measured |
| LMArena Vision: creative writing | 1293 | Not measured | Only Grok 4.5 measured |
| LMArena Document | 1452 | Not measured | Only Grok 4.5 measured |
| Neither model has a result on the other 13 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Grok 4.5 | Qwen3 235B A22B Thinking 2507 | Leader |
|---|---|---|---|
| Neither model has a result on 5 boards here | |||
What each workload costs
Three fixed token mixes at each model's listed price, so per-token rates become a bill.
Chat turn
1K input + 500 output tokens
Qwen3 235B A22B Thinking 2507 costs less
Repository review
50K input + 3K output tokens
Qwen3 235B A22B Thinking 2507 costs less
Agent loop
200K input + 10K output tokens
Qwen3 235B A22B Thinking 2507 costs less
Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.
Specification differences
The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.
| Specification | SpaceXAI Grok 4.5 | Qwen3 235B A22B Thinking 2507 |
|---|---|---|
| Maker | xAI | Qwen |
| Weights | No public weights listed | Open weights |
| Released | July 8, 2026 | July 25, 2025 |
| Context window | 500K | 128K |
| Input $/Mtok | $2.00 | $0.45 |
| Output $/Mtok | $6.00 | $3.50 |
| Coding index | 58.6 | Not benchmarked |
| Independent measurements behind it | 6 | 0 |
| Cost per solved task | $0.62 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Grok 4.5 better than Qwen3 235B A22B Thinking 2507 for coding?
Only Grok 4.5 has a coding score, 58.6 on the Vibe Coding Index; Qwen3 235B A22B Thinking 2507 has none, so the two cannot be ranked on coding. Pick Qwen3 235B A22B Thinking 2507 for output tokens 1.7x cheaper ($3.50 vs $6.00 per million) and work measured by LMArena Text: medicine and healthcare, where it leads.
Which is cheaper, Grok 4.5 or Qwen3 235B A22B Thinking 2507?
Grok 4.5 costs $2.00 per million input tokens and $6.00 per million output tokens; Qwen3 235B A22B Thinking 2507 costs $0.45 and $3.50. Cost per solved task is known for Grok 4.5 only.
Which has the bigger context window, Grok 4.5 or Qwen3 235B A22B Thinking 2507?
Grok 4.5, with 500K tokens against 128K.
How many benchmarks is the Grok 4.5 vs Qwen3 235B A22B Thinking 2507 comparison based on?
43 benchmarks have a result for both models as of October 22, 2026. Grok 4.5's coding index rests on 6 independent benchmarks and Qwen3 235B A22B Thinking 2507's on 0. Where only one model has a result, the page says so instead of naming a winner.