Up to four models
DeepSeek-V4-Pro vs Muse Spark 1.2
On the index (GPT-4 = 100), DeepSeek-V4-Pro scores 163 and Muse Spark 1.2 162. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
163147–178
WeightsOpen
ReleasedAug 2026
Price in / out$0.11 / $4.44
Context1.05M
Core tests taken6 of 9
162146–178
WeightsClosed
ReleasedAug 2026
Price in / out$1.25 / $4.25
Context1.05M
Core tests taken3 of 9
DeepSeek-V4-Pro and Muse Spark 1.2 are too close to call (0 points apart, within the uncertainty), ahead on 2 of 3 shared core tests, and costs 1.7× less.
Test by testthird-party results
Every result as a table
| Test | DeepSeek-V4-Pro | Muse Spark 1.2 |
|---|---|---|
| Reasoning & knowledge | ||
| LiveBench Data Analysis | 79.2% | 76.5% |
| LiveBench IF | 67.7% | 74.3% |
| LiveBench Language | 82.1% | 78.6% |
| LiveBench Reasoning | 85.8% | 90.0% |
| MMLU Pro | 87.0% | 88.3% |
| SimpleQA Verified | 52.9% | 60.3% |
| Maths | ||
| LiveBench Mathematics | 95.1% | 91.2% |
| ProofBench v1.1 | 50.0% | 43.0% |
| Chess puzzles | 46.0% | 27.0% |
| Coding | ||
| Terminal-Bench 4.0 | 14.1% | 6.1% |
| LiveBench Agentic Coding | 54.9% | 57.6% |
| LiveBench Coding | 77.2% | 77.5% |
| Code Migration | 41.5% | 29.9% |
| CyberBench v1.1 | 72.9% | 69.5% |
| IOI | 51.6% | 21.8% |
| ProgramBench | 0.0% | 0.5% |
| SWE-bench | 96.4% | 86.6% |
| Vibe Code Bench v1.1 | 82.3% | 79.1% |
| WeirdML | 66.2% | 60.4% |
| Agents | ||
| APEX-Agents | 47.3% | 45.4% |
| SkillsBench | 53.8% | 53.0% |
| SRE Bench | 1.9% | 0.0% |
| DTBench | 93.9% | 94.7% |
| LMCA | 45.5% | 48.4% |
| Professional work | ||
| Finance Agent (v2) | 50.4% | 60.6% |
| CorpFin v2 | 65.4% | 70.9% |
| EMB | 52.8% | 57.0% |
| LegalBench | 82.4% | 85.3% |
| Legal Research Bench | 40.9% | 43.8% |
| MedCode | 42.5% | 49.4% |
| MedScribe | 80.2% | 90.1% |
| Tax Agent Bench | 26.8% | 32.8% |
| TaxEval v2 | 73.1% | 80.4% |
Tests at least two of these models have taken. Bold rows are the core tests.