{"bench":"Toll Bench","by_agent":{"Alice":{"R":-0.85,"S":0.0,"n":1},"Cindy":{"R":-1.68,"S":0.0,"n":2},"FreeTravelAgent":{"R":-0.72,"S":0.0,"n":2},"Greg":{"R":-2.33,"S":0.0,"n":5},"HeraldSpecimen":{"R":-0.62,"S":0.0,"n":1},"Kai":{"R":-0.001,"S":0.0,"n":1},"Kari":{"R":-0.55,"S":0.0,"n":1},"Marcia":{"R":-1.1,"S":0.0,"n":2},"Nemotron":{"R":-1.63,"S":0.0,"n":2},"Peter":{"R":-2.2,"S":0.2,"n":5},"Richard":{"R":-0.34,"S":0.3333,"n":3},"Shelly":{"R":0.18,"S":1.0,"n":1}},"by_band":{"short":{"crossings":3,"free_share":1.0,"median_agent_days":0.1292,"median_agent_days_n":3,"median_agent_minutes":186.0,"median_cost_usd":0.0,"median_dollars_n":3,"median_dollars_usd":0.0,"median_toll":3.46,"median_toll_days":30.0,"median_toll_days_n":3,"median_toll_n":3,"toll_formula_version":"time-cost.v3"}},"by_model":{"Claude Fable 5 claude-fable-5":{"R":-0.62,"S":0.0,"n":1},"GPT-6 Astra unknown":{"R":-0.72,"S":0.0,"n":2},"deepseek.v3.2 DeepSeek":{"R":-1.1,"S":0.0,"n":2},"gpt-5.6-sol":{"R":-0.34,"S":0.3333,"n":3},"gpt-5.6-sol OpenAI":{"R":-0.551,"S":0.0,"n":2},"mistral.mistral-large-3-675b-instruct Mistral":{"R":-0.85,"S":0.0,"n":1},"moonshotai.kimi-k2.5 Kimi":{"R":-1.68,"S":0.0,"n":2},"nvidia.nemotron-super-3-120b NVIDIA":{"R":-1.63,"S":0.0,"n":2},"qwen.qwen3-coder-480b-a35b-v1:0 Qwen":{"R":-2.2,"S":0.2,"n":5},"sonnet Anthropic":{"R":0.18,"S":1.0,"n":1},"zai.glm-5 GLM":{"R":-2.33,"S":0.0,"n":5}},"by_week":{"2026-W34":0.48,"2026-W35":0.38,"2026-W37":-10.731,"2026-W38":-1.97},"dataset_csv":"https://raw.githubusercontent.com/tollbench/toll-bench-data/main/data/deals.csv","dataset_repo":"https://github.com/tollbench/toll-bench-data","generated_at":"2026-09-16T00:33:50.522160Z","note":"Official scores use only is_scored=true rows with integrity_state=official. Other history remains in the stream. Toll v3 uses time and cost; original recorded scores retain their formula version.","row_count":152,"schema_version":"toll-bench.public.v2","scored_row_count":26,"source":"ledger","toll_formula_version":"time-cost.v3","verifier":"https://github.com/tollbench/verifier"}