Vals Router Index (Vals AI) (Vals Router Index)
We show this table for reference; we do not rank on it.
Vals AI comparison of model routers with single models on long agentic work in finance, coding, law and tax, using the same agents, tasks and graders as the Vals Index.
Evaluation results published by Vals AI. Published leaderboard aggregates and evaluation settings only. Dataset contents are not included.
Accuracy on Vals Router Index — October 20, 2026
We mirror the published accuracy view for Vals Router Index. Gemini 4 Argon leads the public snapshot at 69.24%, followed by Claude Sonnet 5.5 (69.03%) and Muse Wonder Full (66.79%). We do not use these results to rank models overall.
Gemini 4 Argon
Claude Sonnet 5.5
Anthropic
Muse Wonder Full
Meta
27 modelsAgenticCurrentDisplay onlyUpdated October 20, 2026
Accuracy table (27 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Gemini 4 ArgonGoogle | 69.24% | Not reported | Closed |
| 2 | Claude Sonnet 5.5Anthropic | 69.03% | Not reported | Closed |
| 3 | Muse Wonder FullMeta | 66.79% | Not reported | Closed |
| 4 | GPT-6 AstraOpenAI | 65.02% | Not reported | Closed |
| 5 | Muse WonderMeta | 64.88% | Not reported | Closed |
| 6 | Claude Opus 5.5Anthropic | 64.16% | Not reported | Closed |
| 7 | GPT-6.1 SolOpenAI | 63.94% | Not reported | Closed |
| 8 | DeepSeek V4.1 Flash 1017DeepSeek | 61.26% | Not reported | Open |
| 9 | GLM-5.3Z.AI | 59.73% | Not reported | Open |
| 10 | Claude Haiku 5.5Anthropic | 58.71% | Not reported | Closed |
| 11 | Grok 4.7xAI | 58.56% | Not reported | Closed |
| 12 | MiMo-V2.6-ProXiaomi | 57.81% | Not reported | Open |
| 13 | Ramp Routerramp | 57.80% | Not reported | Unknown |
| 14 | Fireworks FireRouterFireworks | 56.46% | Not reported | Unknown |
| 15 | Gemini 3.8 FlashGoogle | 55.95% | Not reported | Closed |
| 16 | TypeSafe Jev Routeropenrouter | 55.65% | Not reported | Unknown |
| 17 | MiMo-V2.6-FlashXiaomi | 54.83% | Not reported | Open |
| 18 | MiniMax M3.1 FlashMiniMax | 53.78% | Not reported | Closed |
| 19 | Hy4 previewTencent | 53.21% | Not reported | Open |
| 20 | DeepSeek V4 Pro 0813DeepSeek | 52.97% | Not reported | Open |
| 21 | Kimi K3Moonshot AI | 52.95% | Not reported | Open |
| 22 | GPT-6 LunaOpenAI | 52.32% | Not reported | Closed |
| 23 | OpenRouter Autoopenrouter | 52.20% | Not reported | Unknown |
| 24 | GLM-5.3-FlashZ.AI | 50.86% | Not reported | Open |
| 25 | Sakana Fugu Maxsakana | 48.11% | Not reported | Unknown |
| 26 | InklingThinking Machines Lab | 28.96% | Not reported | Open |
| 27 | Inkling-SmallThinking Machines Lab | 27.59% | Not reported | Open |
How to read this leaderboard
Operator receipt: 27 sourced rows are currently displayable on this page; the leading published row is Gemini 4 Argon at 69.24%.
Honest limit: Only five routers are measured, each in its default configuration. Fugu Max does not disclose which models it calls, so Vals leaves it out of pool comparisons. Router rows describe a system that selects models per request, not one model.
How Vals Router Index is shown here
We mirror Vals AI's Router Index, updated October 20, 2026. It runs five model routers (OpenRouter Auto, Fireworks FireRouter, Ramp Router, TypeSafe Jev Router and Sakana Fugu Max) through the same agents, tasks and graders as every model on the Vals Index, about 1,250 tasks across eight benchmarks of agentic work in finance, coding, law and tax. Each router ran with its default configuration.
Vals compares each router with single models on its Index and reports accuracy, cost per test and duration. The board lists the five routers beside the single models in Vals' comparison; a router picks models per request, so its row is a system rather than one model. The results stay display only and do not enter model rankings.
Snapshot
The published Vals Router Index snapshot places Gemini 4 Argon first at 69.24%. The third row is 2.45 points behind. The broader top-10 range is 10.53 points, so the table still separates the published systems.
27 models have been evaluated on Vals Router Index. The benchmark falls in the Agentic category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. Vals Router Index is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About Vals Router Index
Year
2026
Tasks
About 1,250 agentic tasks
Format
Accuracy, cost per test and duration
Difficulty
Long tool-using agent work in finance, coding, law and tax
Vals published the index on October 20, 2026. Five routers (OpenRouter Auto, Fireworks FireRouter, Ramp Router, TypeSafe Jev Router and Sakana Fugu Max) run about 1,250 tasks from the eight agentic benchmarks of the Vals Index, six of them private, each in its default configuration. Vals reports accuracy, cost per test and duration and compares each router with single models. We mirror the published aggregates as a display-only board, outside overall and category rankings.
Freshness and provenance
Version
Vals Router Index 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does Vals Router Index measure?
Vals AI comparison of model routers with single models on long agentic work in finance, coding, law and tax, using the same agents, tasks and graders as the Vals Index.
Which model leads the published Vals Router Index snapshot?
Gemini 4 Argon currently leads the published Vals Router Index snapshot with 69.24% accuracy. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on Vals Router Index?
The October 20, 2026 snapshot contains 27 AI models.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 6,100+ subscribers.
One email each week. Unsubscribe anytime.