Model comparison
DeepSeek V4 Pro vs Grok 4.20 (Non-Reasoning)
DeepSeek V4 Pro is the stronger model overall, scoring 54.3 to 48.6 on the Noometry Index.
Last verified . 38 shared benchmarks.
DeepSeekSummary
- They share 38 benchmarks with published results for both. DeepSeek V4 Pro scores higher in 5 categories and Grok 4.20 (Non-Reasoning) in 4 categories; 6 gaps are clear of the uncertainty.
- The widest gap is in math, where DeepSeek V4 Pro leads 64.8 to 48.2.
- The biggest single-benchmark swing is ProofBench: 50% for DeepSeek V4 Pro and 14% for Grok 4.20 (Non-Reasoning).
- DeepSeek V4 Pro is cheaper at $0.66 / $1.98 per million input/output tokens, against $1.25 / $2.50 for Grok 4.20 (Non-Reasoning).
- DeepSeek V4 Pro has downloadable open weights; the other is API-only.
Side by side
| DeepSeek V4 Pro | Grok 4.20 (Non-Reasoning) | |
|---|---|---|
| Provider | DeepSeek | xAI |
| Noometry Index | 54.3 | 48.6 |
| Released | 2026-04-24 | 2026-02-17 |
| Weights | Open | Proprietary |
| Context window | 1M | 1M |
| Max output | 393K | 30K |
| Input $ / M tokens | $0.66 | $1.25 |
| Output $ / M tokens | $1.98 | $2.50 |
| Results tracked | 48 | 46 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek V4 Pro leads
DeepSeek V4 Pro: 52.4 (#34), Grok 4.20 (Non-Reasoning): 42.1 (#112)
| Benchmark | DeepSeek V4 Pro | Grok 4.20 (Non-Reasoning) |
|---|---|---|
| LMArena WebDev | 1582 | 1375 |
| WeirdML | 66.2% | 52.3% |
| LMArena Coding | 1470 | 1459 |
| ALE-Bench | 1,403 | 1,150 |
| SWE-bench Verified | 77.6% | — |
| FrontierCode | 28.6% | — |
| SciCode | 51% | — |
Agentic & Tool Use Grok 4.20 (Non-Reasoning) leads
DeepSeek V4 Pro: 32.8 (#58), Grok 4.20 (Non-Reasoning): 34.4 (#46)
| Benchmark | DeepSeek V4 Pro | Grok 4.20 (Non-Reasoning) |
|---|---|---|
| Vending-Bench 2 | 3,285 | 4,663 |
| Terminal-Bench | — | 57.3% |
| APEX-Agents | 47.3% | — |
| τ²-bench Banking | — | 18% |
| LMArena Search | — | 1189 |
Reasoning DeepSeek V4 Pro leads
DeepSeek V4 Pro: 56.5 (#24), Grok 4.20 (Non-Reasoning): 52.3 (#32)
| Benchmark | DeepSeek V4 Pro | Grok 4.20 (Non-Reasoning) |
|---|---|---|
| ARC-AGI-2 | 61.3% | 65.1% |
| Kagi LLM Benchmark | 53.5% | 75% |
| NYT Connections (extended) | 91.3% | 85.4% |
| ARC-AGI-1 | 90.5% | 89.5% |
| Chess Puzzles | 47% | 24% |
| LMArena Hard Prompts | 1461 | 1451 |
| DTBench | 93.9% | 90.1% |
| LMCA | 45.5% | 38.7% |
| Epoch Capabilities Index | 155.31 | 151.98 |
| ForecastBench | 56.1 | 61.4 |
| CritPt | 18% | — |
| Thematic Generalization | — | 63.8% |
| Mystery Game Puzzles | 43% | — |
| Surface Evolver Bench | 40% | — |
Math DeepSeek V4 Pro leads
DeepSeek V4 Pro: 64.8 (#30), Grok 4.20 (Non-Reasoning): 48.2 (#65)
| Benchmark | DeepSeek V4 Pro | Grok 4.20 (Non-Reasoning) |
|---|---|---|
| FrontierMath (Tiers 1-3) | 64.6% | 44.9% |
| FrontierMath Tier 4 | 26.8% | 17.1% |
| OTIS Mock AIME 2024-2025 | 98.6% | 92.2% |
| ProofBench | 50% | 14% |
| LMArena Math | 1455 | 1455 |
| MathArena Final-Answer Competitions | 76.6% | — |
Knowledge DeepSeek V4 Pro leads
DeepSeek V4 Pro: 59.5 (#31), Grok 4.20 (Non-Reasoning): 52.8 (#60)
| Benchmark | DeepSeek V4 Pro | Grok 4.20 (Non-Reasoning) |
|---|---|---|
| GPQA Diamond | 91.7% | 89.3% |
| SimpleQA Verified | 52.9% | 30.2% |
| LMArena Expert | 1464 | 1439 |
| Vectara Hallucination Rate | 8.6% | — |
Multimodal Not comparable
DeepSeek V4 Pro: —, Grok 4.20 (Non-Reasoning): 33.3 (#98)
| Benchmark | DeepSeek V4 Pro | Grok 4.20 (Non-Reasoning) |
|---|---|---|
| LMArena Vision | — | 1263 |
| Blueprint-Bench 2 | — | 0% |
| LMArena Document | — | 1416 |
Multilingual Too close to call
DeepSeek V4 Pro: 54.4 (#45), Grok 4.20 (Non-Reasoning): 54.5 (#40)
| Benchmark | DeepSeek V4 Pro | Grok 4.20 (Non-Reasoning) |
|---|---|---|
| LMArena Non-English | 1439 | 1441 |
| LMArena Chinese | 1486 | 1481 |
| LMArena French | 1472 | 1476 |
| LMArena German | 1458 | 1465 |
| LMArena Japanese | 1445 | 1449 |
| LMArena Korean | 1447 | 1417 |
| LMArena Russian | 1453 | 1458 |
| LMArena Spanish | 1458 | 1443 |
Instruction Following DeepSeek V4 Pro leads
DeepSeek V4 Pro: 76.1 (#47), Grok 4.20 (Non-Reasoning): 74.8 (#83)
| Benchmark | DeepSeek V4 Pro | Grok 4.20 (Non-Reasoning) |
|---|---|---|
| LMArena Instruction Following | 1448 | 1420 |
Long Context Too close to call
DeepSeek V4 Pro: 45.0 (#51), Grok 4.20 (Non-Reasoning): 45.5 (#34)
| Benchmark | DeepSeek V4 Pro | Grok 4.20 (Non-Reasoning) |
|---|---|---|
| CL-bench Life | 13.5% | 11.9% |
| LMArena Longer Query | 1458 | 1437 |
| CL-bench | — | 22.2% |
Writing & Preference Too close to call
DeepSeek V4 Pro: 65.5 (#46), Grok 4.20 (Non-Reasoning): 65.7 (#44)
| Benchmark | DeepSeek V4 Pro | Grok 4.20 (Non-Reasoning) |
|---|---|---|
| LMArena Text | 1451 | 1451 |
| LMArena Creative Writing | 1446 | 1438 |
| EQ-Bench Creative Writing | 1553 | 1574 |
| LMArena Multi-Turn | 1467 | 1456 |
| EQ-Bench 4 | 1166 | — |
Frequently asked questions
Is DeepSeek V4 Pro better than Grok 4.20 (Non-Reasoning)?
DeepSeek V4 Pro is the stronger model overall, scoring 54.3 to 48.6 on the Noometry Index.
Which is cheaper, DeepSeek V4 Pro or Grok 4.20 (Non-Reasoning)?
DeepSeek V4 Pro is cheaper. It lists at $0.66 per million input tokens and $1.98 per million output tokens; Grok 4.20 (Non-Reasoning) lists at $1.25 and $2.50.
Is DeepSeek V4 Pro or Grok 4.20 (Non-Reasoning) better for coding?
DeepSeek V4 Pro scores higher on coding benchmarks: 52.4 versus 42.1 in the Noometry coding category.
Which has the bigger context window?
Both accept 1M tokens.
How many benchmarks do DeepSeek V4 Pro and Grok 4.20 (Non-Reasoning) share?
38 benchmarks have published results for both models. DeepSeek V4 Pro has 48 scored results on Noometry and Grok 4.20 (Non-Reasoning) has 46.