𝜏³-Banking

Un benchmark de support client fintech issu du framework 𝜏-Knowledge, qui évalue si les agents peuvent naviguer dans une grande base de connaissances non structurée et exécuter des appels d'outils en plusieurs étapes pour résoudre des workflows bancaires réalistes.

Modèles: 89 actuel, 210 au total
Médiane: 17.5%
Plage: 2.9%–50.3%

Distribution des scores

Le plus bas: 2.9%
Médiane: 17.5%
Le plus élevé: 50.3%

Classement

Inclure les modèles obsolètes

Rang	Modèle	Score
1	GLM-5.3 (Max) Z AI	50.3%
2	Qwen3.8 2.4T A95B Alibaba	49.1%
3	Qwen3.8 27B (Xhigh) Alibaba	48.0%
4	Qwen3.8 Max (0902) Alibaba	47.8%
5	Qwen3.8 27B (Medium) Alibaba	47.4%
6	GLM-5.3-Flash Z AI	47.2%
7	Gemini 3.8 Flash (Medium) Google	45.8%
8	Qwen3.8-Flash-Next Alibaba	45.4%
9	Gemini 3.8 Flash (High) Google	44.9%
10	GPT-6 Astra (Xhigh) OpenAI	43.1%
11	GPT-6 Astra (Max) OpenAI	41.4%
12	DeepSeek V4 Flash Vision (Max) DeepSeek	41.0%
13	Inkling 2 Small Thinking Machines	40.4%
14	GPT-6 Astra (High) OpenAI	40.0%
15	DeepSeek V4 Pro 0813 (Max) DeepSeek	39.6%
16	Ling-3.0-flash-Fin InclusionAI	38.6%
17	Agnes 2.5 Pro Beta Sapiens AI	35.7%
18	GPT-6 Astra (Medium) OpenAI	35.5%
19	Motif 3 Motif Technologies	35.3%
20	Ling-3.0-flash-VL InclusionAI	34.4%
21	K2 Horizon 375B A23B Institute of Foundation Models	34.2%
22	Gemini 3.8 Flash (Low) Google	33.2%
23	Qwen3.8 27B (Low) Alibaba	32.2%
24	GPT-6 Astra (Low) OpenAI	32.0%
25	K2 Horizon MoVA 36B A4B Institute of Foundation Models	32.0%
26	Inkling (Xhigh) Thinking Machines	29.1%
27	Quasar 438B (Max, Based on GLM-5.2) Multiverse Computing	28.7%
28	K2 Horizon 7B Institute of Foundation Models	27.6%
29	Apodex 1.1 Apodex	25.2%
30	Muse Glimmer (High) Meta	23.5%
31	Solar Pro 4 Upstage	23.3%
32	Hy3 Tencent	22.9%
33	G9v3-39A5B AI9Stars	22.1%
34	Solar Open2 250B Upstage	21.6%
35	Gemini 3.1 Pro Preview Google	21.4%
36	Ling 3.0 Tiny InclusionAI	20.8%
37	MiniCPM5-2B OpenBMB	20.8%
38	Kimi K2.7 Code Kimi	20.2%
39	Qwen3.8 27B (Non-reasoning) Alibaba	20.0%
40	K2 Horizon 3.7B Institute of Foundation Models	19.6%
41	Inkling Small Thinking Machines	18.8%
42	Ring-2.6-1T InclusionAI	17.9%
43	Gemini 3.5 Flash-Lite Google	17.5%
44	Nex-N2-Pro (Based on Qwen3.5-397B-A17B) Nex AGI	17.5%
45	Qwen3.7 Plus Alibaba	17.5%
46	A.X-K2 SK Telecom	16.1%
47	MiniMax-M3 MiniMax	15.3%
48	Qwen3.5 122B A10B (Reasoning) Alibaba	15.3%
49	Mistral Medium 3.5 Mistral	15.1%
50	Gemma 4 31B (Reasoning) Google	14.8%