K2 Horizon MoVA Card - tau3-Banking: leaderboard
Agentic tool use in the banking domain of tau3-bench: the model acts as a customer-service agent, calling tools and following domain policy in a conversation with a simulated user. Scores in percent from IFM's K2 Horizon MoVA 36B A4B model card, which states that all reported MoVA results use high reasoning effort. The card takes its six baseline scores from Artificial Analysis (Muse Glimmer-30B at high reasoning effort, the other open models in reasoning mode), so this table mixes IFM's MoVA result with copied third-party numbers and is separate from Artificial Analysis's own tau3-Banking board.
Metric: Score (%). Source: huggingface.co. Saturation forecast: Around September 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Muse Glimmer 30B (High) | 23.5 | #371 (Muse Glimmer 30B) |
| 2 | Gemma 4 31B (IT) (Reasoning) | 14.8 | #343 (Gemma 4 31B (IT)) |
| 3 | Qwen 3.6 35B A3B (Reasoning) | 9.3 | #461 (Qwen 3.6 35B A3B) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=k2-horizon-mova-card-tau3-banking · How It Works · Data refreshed daily, snapshot 2026-11-12.