Best LLMs for Tool Calling — Qwen

Match · Updated August 2026

The best Qwen model for tool calling is Qwen3.6 Plus, completing 97.7% of Tau²-Bench's multi-turn tool-use tasks. Qwen3.6 Max Preview (95.9%) and Qwen3.5 397B A17B (95.6%) round out the top three.

97.7%τ²-Bench

39.6Intelligence

55 t/sSpeed

$0.325Input /M

1MContext

1
qwen3.6-plus
ReasoningToolsJSON+139.6 intel · $0.325/M · 55 t/s
97.7%
τ²-Bench
2
qwen3.6-max-preview
ReasoningToolsJSON40.0 intel · $1.03/M · 262K ctx
95.9%
τ²-Bench
3
qwen3.5-397b-a17b
ReasoningToolsJSON+133.7 intel · $0.390/M · 69 t/s
95.6%
τ²-Bench
4
qwen3.6-35b-a3b
ReasoningToolsJSON+131.6 intel · $0.140/M · 158 t/s
95.3%
τ²-Bench
5
qwen3.7-max
ReasoningToolsJSON46.0 intel · $1.48/M · 207 t/s
94.7%
τ²-Bench
6
qwen3.6-27b
ReasoningToolsJSON+137.1 intel · $0.289/M · 60 t/s
94.2%
τ²-Bench
7
qwen3.5-27b
ReasoningToolsJSON+133.8 intel · $0.195/M · 262K ctx
93.9%
τ²-Bench
8
qwen3.5-122b-a10b
ReasoningToolsJSON+132.3 intel · $0.260/M · 127 t/s
93.6%
τ²-Bench
9
qwen3.7-plus
ReasoningToolsJSON+139.0 intel · $0.320/M · 12 t/s
93.0%
τ²-Bench
10
qwen3.5-35b-a3b
ReasoningToolsJSON+129.3 intel · $0.140/M · 262K ctx
89.2%
τ²-Bench
11
qwen3.5-9b
ReasoningToolsJSON+121.4 intel · $0.100/M · 74 t/s
86.8%
τ²-Bench
12
qwen3-max-thinking
ReasoningToolsJSON31.7 intel · $0.780/M · 262K ctx
83.6%
τ²-Bench
13
qwen3-coder-next
ToolsJSON21.1 intel · $0.120/M · 133 t/s
79.5%
τ²-Bench
14
qwen3-max
ToolsJSON24.0 intel · $0.780/M · 29 t/s
74.3%
τ²-Bench
15
qwen3-vl-235b-a22b-instruct
ToolsJSONVision14.3 intel · $0.210/M · 19 t/s
35.1%
τ²-Bench
16
qwen3-coder-30b-a3b-instruct
ToolsJSON13.6 intel · $0.070/M · 81 t/s
34.5%
τ²-Bench
17
qwen-2.5-72b-instruct
ToolsJSON9.6 intel · $0.360/M · 33K ctx
34.5%
τ²-Bench
18
qwen3-vl-32b-instruct
ToolsJSONVision11.1 intel · $0.104/M · 38 t/s
29.2%
τ²-Bench
19
qwen3-vl-8b-instruct
ToolsJSONVision8.4 intel · $0.117/M · 42 t/s
29.2%
τ²-Bench
20
qwen3-next-80b-a3b-instruct
ToolsJSON13.7 intel · $0.090/M · 74 t/s
21.6%
τ²-Bench
21
qwen3-vl-30b-a3b-instruct
ToolsJSONVision10.0 intel · $0.150/M · 34 t/s
19.0%
τ²-Bench

How this is ranked

AI models ranked by Tau²-Bench — multi-turn conversations where the model has to call the right tools, in the right order, against a real API to complete a customer task. This measures whether function calling actually works under pressure, which is a different question from whether a model supports the parameter at all.

Frequently asked

Which Qwen model is best at tool calling?

The best Qwen model for tool calling is Qwen3.6 Plus, completing 97.7% of Tau²-Bench's multi-turn tool-use tasks. Qwen3.6 Max Preview (95.9%) and Qwen3.5 397B A17B (95.6%) round out the top three.

What's a good alternative to Qwen3.6 Plus?

Qwen3.6 Max Preview (95.9%) is the closest alternative on this metric, followed by Qwen3.5 397B A17B (95.6%). See the full ranking above for the tradeoffs.

How many Qwen models are there?

modelgrep tracks 49 Qwen models with live benchmarks, speed, latency and per-provider pricing, led on intelligence by Qwen3.7 Max. 21 of them qualify for this ranking.