Model comparison
DeepSeek-V3.2-Exp vs Grok 4.5
Grok 4.5 is the stronger model overall, scoring 55.0 to 44.9 on the Noometry Index. DeepSeek-V3.2-Exp costs 10× less per token, which makes it the better buy when Grok 4.5's lead doesn't matter for your workload.
Last verified . 34 shared benchmarks.
DeepSeekSummary
- They share 34 benchmarks with published results for both. DeepSeek-V3.2-Exp scores higher in 1 category and Grok 4.5 in 8 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where Grok 4.5 leads 56.0 to 22.0.
- The biggest single-benchmark swing is ARC-AGI-2: 4% for DeepSeek-V3.2-Exp and 52.6% for Grok 4.5.
- DeepSeek-V3.2-Exp is cheaper at $0.26 / $0.38 per million input/output tokens, against $2 / $6 for Grok 4.5.
- Grok 4.5 accepts more context: 500K tokens versus 164K.
- DeepSeek-V3.2-Exp has downloadable open weights; the other is API-only.
Side by side
| DeepSeek-V3.2-Exp | Grok 4.5 | |
|---|---|---|
| Provider | DeepSeek | xAI |
| Noometry Index | 44.9 | 55.0 |
| Released | 2025-09-29 | 2026-07-08 |
| Weights | Open | Proprietary |
| Context window | 164K | 500K |
| Max output | 66K | 500K |
| Input $ / M tokens | $0.26 | $2 |
| Output $ / M tokens | $0.38 | $6 |
| Results tracked | 48 | 52 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Grok 4.5 leads
DeepSeek-V3.2-Exp: 46.4 (#61), Grok 4.5: 51.7 (#34)
| Benchmark | DeepSeek-V3.2-Exp | Grok 4.5 |
|---|---|---|
| LMArena WebDev | 1362 | 1552 |
| SciCode | 38.9% | 54.1% |
| WeirdML | 39.5% | 46.4% |
| LMArena Coding | 1454 | 1473 |
| DeepSWE | — | 53.8% |
| FrontierCode | — | 42.4% |
| SWE-bench Verified (bash only) | 70% | — |
| Aider Polyglot | 74.2% | — |
| SWE-bench Multilingual | 59% | — |
| ALE-Bench | — | 1,309 |
Agentic & Tool Use Grok 4.5 leads
DeepSeek-V3.2-Exp: 37.9 (#31), Grok 4.5: 44.4 (#15)
| Benchmark | DeepSeek-V3.2-Exp | Grok 4.5 |
|---|---|---|
| Vending-Bench 2 | 1,034 | 3,887 |
| Terminal-Bench | 39.6% | — |
| APEX-Agents | — | 56.2% |
| Berkeley Function Calling Leaderboard | 56.7% | — |
| TheAgentCompany | 42.9% | — |
| τ²-bench Banking | — | 47.9% |
| PostTrainBench | — | 23.4% |
| GBAEval | — | 65.4% |
| GDP.pdf | — | 14% |
| LMArena Search | — | 1213 |
Reasoning Grok 4.5 leads
DeepSeek-V3.2-Exp: 22.0 (#185), Grok 4.5: 56.0 (#23)
| Benchmark | DeepSeek-V3.2-Exp | Grok 4.5 |
|---|---|---|
| ARC-AGI-2 | 4% | 52.6% |
| Kagi LLM Benchmark | 52.2% | 83.5% |
| NYT Connections (extended) | 36.7% | 79.9% |
| ARC-AGI-1 | 57% | 87.2% |
| CritPt | 2.9% | 15.4% |
| Chess Puzzles | 14% | 36% |
| LMArena Hard Prompts | 1433 | 1462 |
| DTBench | 87.7% | 96.5% |
| LMCA | 29.1% | 45.2% |
| Epoch Capabilities Index | 146.28 | 154.02 |
| SimpleBench | — | 70% |
| Thematic Generalization | 65% | — |
| Surface Evolver Bench | — | 74.4% |
Math Grok 4.5 leads
DeepSeek-V3.2-Exp: 40.8 (#75), Grok 4.5: 60.7 (#29)
| Benchmark | DeepSeek-V3.2-Exp | Grok 4.5 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 87.8% | 97.8% |
| ProofBench | 8% | 31% |
| LMArena Math | 1435 | 1457 |
| FrontierMath (Tiers 1-3) | — | 57.2% |
| FrontierMath Tier 4 | — | 24.4% |
| MathArena Final-Answer Competitions | 57.7% | — |
| FrontierMath (Feb 2025 set) | 22.1% | — |
| FrontierMath Tier 4 (v1) | 2.1% | — |
Knowledge Grok 4.5 leads
DeepSeek-V3.2-Exp: 51.6 (#61), Grok 4.5: 62.3 (#21)
| Benchmark | DeepSeek-V3.2-Exp | Grok 4.5 |
|---|---|---|
| GPQA Diamond | 83.4% | 93.4% |
| LMArena Expert | 1436 | 1466 |
| SimpleQA Verified | — | 48.3% |
| Vectara Hallucination Rate | 5.3% | — |
Multimodal Not comparable
DeepSeek-V3.2-Exp: —, Grok 4.5: 37.8 (#63)
| Benchmark | DeepSeek-V3.2-Exp | Grok 4.5 |
|---|---|---|
| LMArena Vision | — | 1287 |
| Blueprint-Bench 2 | — | 27.3% |
| Furniture Assembly | — | 22.5% |
| LMArena Document | — | 1452 |
Multilingual Grok 4.5 leads
DeepSeek-V3.2-Exp: 52.4 (#85), Grok 4.5: 54.5 (#43)
| Benchmark | DeepSeek-V3.2-Exp | Grok 4.5 |
|---|---|---|
| LMArena Non-English | 1409 | 1439 |
| LMArena Chinese | 1461 | 1497 |
| LMArena French | 1435 | 1458 |
| LMArena German | 1440 | 1448 |
| LMArena Japanese | 1377 | 1425 |
| LMArena Korean | 1372 | 1402 |
| LMArena Russian | 1423 | 1447 |
| LMArena Spanish | 1441 | 1452 |
Instruction Following Grok 4.5 leads
DeepSeek-V3.2-Exp: 74.7 (#89), Grok 4.5: 76.3 (#45)
| Benchmark | DeepSeek-V3.2-Exp | Grok 4.5 |
|---|---|---|
| LMArena Instruction Following | 1413 | 1446 |
Long Context DeepSeek-V3.2-Exp leads
DeepSeek-V3.2-Exp: 47.6 (#15), Grok 4.5: 44.8 (#54)
| Benchmark | DeepSeek-V3.2-Exp | Grok 4.5 |
|---|---|---|
| LMArena Longer Query | 1428 | 1463 |
| Fiction.LiveBench | 83.3% | — |
| CL-bench | 13.2% | — |
| CL-bench Life | 9.5% | — |
Writing & Preference Grok 4.5 leads
DeepSeek-V3.2-Exp: 62.5 (#73), Grok 4.5: 65.9 (#40)
| Benchmark | DeepSeek-V3.2-Exp | Grok 4.5 |
|---|---|---|
| LMArena Text | 1425 | 1448 |
| LMArena Creative Writing | 1403 | 1442 |
| EQ-Bench Creative Writing | 1515 | 1579 |
| LMArena Multi-Turn | 1427 | 1457 |
Frequently asked questions
Is DeepSeek-V3.2-Exp better than Grok 4.5?
Grok 4.5 is the stronger model overall, scoring 55.0 to 44.9 on the Noometry Index. DeepSeek-V3.2-Exp costs 10× less per token, which makes it the better buy when Grok 4.5's lead doesn't matter for your workload.
Which is cheaper, DeepSeek-V3.2-Exp or Grok 4.5?
DeepSeek-V3.2-Exp is cheaper. It lists at $0.26 per million input tokens and $0.38 per million output tokens; Grok 4.5 lists at $2 and $6.
Is DeepSeek-V3.2-Exp or Grok 4.5 better for coding?
Grok 4.5 scores higher on coding benchmarks: 51.7 versus 46.4 in the Noometry coding category.
Which has the bigger context window?
Grok 4.5 does, with 500K tokens against 164K.
How many benchmarks do DeepSeek-V3.2-Exp and Grok 4.5 share?
34 benchmarks have published results for both models. DeepSeek-V3.2-Exp has 48 scored results on Noometry and Grok 4.5 has 52.