Vals Index v2.1
Indexes · 2026-09-25
Vals Index v2.1 combines GDP-weighted sector means: finance (Finance Agent v2 and EMB, weight 8.0), coding (Terminal-Bench 4.0, Vibe Code Bench and the fixed Code Migration subset, weight 5.6), legal (Legal Research and HLAB, weight 1.2), and Tax Agent Bench (weight 0.5), divided by 15.3. It replaces Terminal-Bench 2.1 and adds tax relative to v2. Component runs use their respective harnesses. Mixed-model fallback aggregates are excluded from single-model observations.
Top models (higher is better)
| Model | Score |
|---|---|
| Muse Wonder Full | 64.6 |
| GPT-6.1 Sol | 61.2 |
| Muse Wonder | 61.1 |
| Muse Spark 1.3 | 58.2 |
| GPT-5.6 Sol | 58.0 |
| GPT-6 Sol | 57.5 |
| DeepSeek V4.1 Flash 1017 | 56.6 |
| MiMo-V2.6-Pro | 55.2 |
| Opus 4.8 | 55.1 |
| Grok 4.7 | 54.9 |
| Gemini 3.8 Flash | 54.9 |
| Claude Haiku 5.5 | 54.4 |
| GLM-5.3 | 53.5 |
| MiMo-V2.6-Flash | 53.2 |
| GPT-5.6 Terra | 53.1 |
| Grok 4.6 | 52.1 |
| Sonnet 5 | 51.9 |
| GPT-5.6 Luna | 51.7 |
| Gemini 3.7 Flash | 51.3 |
| GPT-6 Luna | 51.2 |
| Ember-1 | 50.9 |
| Kimi K3 | 50.3 |
| Hy4 Preview | 49.9 |
| Step 5 Preview | 49.4 |
| Muse Spark 1.2 | 49.3 |
| Mistral Large 4 Preview | 48.7 |
| Qwen3.8-Max | 48.3 |
| DeepSeek-V4-Flash-0731 | 48.0 |
| DeepSeek V4 Pro 0813 | 47.6 |
| Muse Spark Open | 46.6 |
| Gemini 3.5 Flash | 44.8 |
| Grok 4.5 | 44.7 |
| GLM-5.3 Flash | 44.6 |
| DeepSeek-V4-Pro | 38.6 |
| MiniMax M3 | 36.5 |
| MiMo-V2.5-Pro | 33.8 |
| Gemini 3.1 Pro Preview | 33.4 |
| GPT-5.4 Mini | 33.2 |
| Inkling | 28.7 |
| Inkling-Small | 25.5 |