← Back to leaderboard

gemma4:26b

17 GB measured VRAM · 12GB+ tier · tool-gauntlet-v1
Tier 1 — floor 5/5
seed 1 (passed): correct single call
131 tokens · 13.329s
assistant
get_weather({'city': 'Seattle'})
Tier 2 — restraint 5/5
seed 1 (passed): correctly answered directly, no spurious tool call
45 tokens · 0.677s
assistant
The capital of Japan is Tokyo.
Tier 3 — selection 5/5
seed 1 (passed): correctly selected get_time among 3 plausible tools
122 tokens · 1.304s
assistant
get_time({'city': 'Tokyo'})
Tier 4 — extraction 5/5
seed 1 (passed): correctly extracted and typed all fields from messy prose
186 tokens · 1.832s
assistant
book_flight({'date': '2026-03-24', 'destination': 'Denver', 'origin': 'Chicago', 'passengers': 3})
Tier 5 — chaining 5/5
seed 1 (passed): correctly chained: location lookup, then weather for that result
184 tokens · 2.062s
assistant
get_user_location({})
assistant
get_weather({'city': 'Seattle'})
Tier 6 — strict schema 5/5
seed 1 (passed): correctly extracted enum, array, and date under full schema constraints
194 tokens · 1.853s
assistant
create_calendar_event({'attendees': ['Maria', 'James'], 'date': '2026-04-02', 'priority': 'high', 'title': 'Budget Review'})