Benchmark leaderboard
τ²-Bench Telecom measures whether an agent can resolve telecom customer-service tasks through tool calls while a simulated user talks back. Artificial Analysis runs every model through the same harness, so the pass rate compares models rather than scaffolds. This page lists every model ModelCap tracks with a published AA τ²-Bench Telecom result, its best evaluated configuration, the source's own rank, and the model's ModelCap Index position, API price and context window.
Snapshot as of 30 September 2026
Gemini 3.1 Pro Preview from Google leads the τ²-Bench Telecom (Artificial Analysis run) among the 46 tracked models with 95.6% (Reasoning configuration), per the source snapshot published 30 September 2026.
Sorted by published score, one row per model. Source: Artificial Analysis.
AA τ²-Bench Telecom ranks models by τ²-Bench Telecom task pass rate. ModelCap ingests the board as published, matches each entry to a catalogue model with a reviewed identity, and shows the source's score, interval and rank unchanged. AA τ²-Bench Telecom is displayed as a public reference board and does not contribute to the ModelCap Index. Its published measurements and tested configurations remain visible here. The methodology page lists the sources admitted to the Index. Read the methodology and identity rules.
τ²-Bench Telecom measures whether an agent can resolve telecom customer-service tasks through tool calls while a simulated user talks back. Artificial Analysis runs every model through the same harness, so the pass rate compares models rather than scaffolds. It is published by Artificial Analysis.
Gemini 3.1 Pro Preview (Google) holds the top AA τ²-Bench Telecom score among the models ModelCap tracks, at 95.6% as of the source snapshot published 30 September 2026.
46 tracked models have a published AA τ²-Bench Telecom result on ModelCap; the source board itself lists 152 entries. Each row shows the model's best evaluated configuration.
Qwen3.5 397B A17B from Qwen is the highest-scoring model with openly downloadable weights on this board, at 95.6%.
Among the ten highest-scoring priced models, Qwen3.5-9B has the lowest listed output price at $0.15 per 1M tokens while scoring 86.8%.
The board ranks models by τ²-Bench Telecom task pass rate; higher scores are better. ModelCap shows the source's own score, interval and rank and never re-runs the evaluation.
AA τ²-Bench Telecom is displayed as a public reference board and does not contribute to the ModelCap Index. Its published measurements and tested configurations remain visible here. The methodology page lists the sources admitted to the Index.
The newest AA τ²-Bench Telecom publication ModelCap holds is dated 30 September 2026. This page uses the published snapshot; source collection, publication and page updates happen separately.