korshunov.ai Everything that matters in ML
EN
Benchmark · agentic

Berkeley Function-Calling Leaderboard

6 results 5 models

Berkeley's function/tool-calling accuracy leaderboard.

0 24 48 72 96 2024-12-17 2025-10-15 2026-08-13 Falcon3-10B-Instruct · 86.3 · 2024-12-17 GLM-4.5 · 90.6 · 2025-08-06 GLM-4.5-Air · 76.4 · 2025-08-06 Pokee-Isaac 28B · 70.9 · 2026-08-08 LFM2.5-VL-3B · 32.5 · 2026-08-12 LFM2.5-VL-3B · 32.5 · 2026-08-13
Falcon3-10B-Instruct GLM-4.5 GLM-4.5-Air Pokee-Isaac 28B LFM2.5-VL-3B
Timeline
Date Model Score Source
2026-08-13 LFM2.5-VL-3B 32.5% Liquid AI releases LFM2.5-VL-3B, a 3B on-device vision-language model with tool calling
2026-08-12 LFM2.5-VL-3B 32.5% LFM2.5-VL-3B improves screen understanding and function calling for edge deployment
2026-08-08 Pokee-Isaac 28B 70.94% Pokee AI releases Pokee-Isaac 28B, a 10M-token context model for in-boundary deployment
2025-08-06 GLM-4.5 90.6% Zhipu AI releases GLM-4.5 models matching Claude and DeepSeek
2025-08-06 GLM-4.5-Air 76.4% Zhipu AI releases GLM-4.5 models matching Claude and DeepSeek
2024-12-17 Falcon3-10B-Instruct 86.3% Falcon3 family releases five open models with improved science, math, and code capabilities