Function Calling on Berkeley Function Calling Leaderboard (BFCL) v4
77.25Simple AccuracyQWEN3-32B
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| QWEN3-32B2026.04 | 77.25 | 96 | 93 | 91.5 | 89.44 | 81.87 | 85.66 | 80 | |
| GOEDEL-PROVER-V2-RL + 100 agentic SFTStarting Checkpoint=GOEDEL-PROVER-V2-32B-RL, Agentic SFT Training Data Volume=1002026.04 | 77.25 | 94 | 88 | 84 | 85.81 | 69.2 | 77.51 | 21.67 | |
| GOEDEL-PROVER-V2-SFT + 1k agentic SFTStarting Checkpoint=GOEDEL-PROVER-V2-32B-SFT, Agentic SFT Training Data Volume=1k2026.04 | 75 | 94 | 85.5 | 79.5 | 83.5 | 72.91 | 78.21 | 36.25 | |
| GOEDEL-PROVER-V2-SFT + 100 agentic SFTStarting Checkpoint=GOEDEL-PROVER-V2-32B-SFT, Agentic SFT Training Data Volume=1002026.04 | 74.83 | 92.5 | 86.5 | 81.5 | 83.83 | 72.91 | 78.37 | 21.67 | |
| GOEDEL-PROVER-V2-RL + 1k agentic SFTStarting Checkpoint=GOEDEL-PROVER-V2-32B-RL, Agentic SFT Training Data Volume=1k2026.04 | 73.83 | 92 | 89.5 | 87.5 | 85.71 | 74.91 | 80.31 | 33.75 | |
| GOEDEL-PROVER-V2-RL + 18k agentic SFTStarting Checkpoint=GOEDEL-PROVER-V2-32B-RL, Agentic SFT Training Data Volume=18k2026.04 | 71.5 | 94 | 88.5 | 81.5 | 83.88 | 73.64 | 78.76 | 34.58 | |
| GOEDEL-PROVER-V2-SFT + 18k agentic SFTStarting Checkpoint=GOEDEL-PROVER-V2-32B-SFT, Agentic SFT Training Data Volume=18k2026.04 | 69.33 | 88.5 | 82.5 | 80.5 | 80.21 | 70.32 | 75.27 | 49.17 | |
| GOEDEL-PROVER-V2-RLStarting Checkpoint=GOEDEL-PROVER-V2-32B-RL, Agentic SFT Training Data Volume=02026.04 | 30.5 | 25 | 13.5 | 3.5 | 18.13 | 18.73 | 18.43 | 91.25 | |
| GOEDEL-PROVER-V2-SFTStarting Checkpoint=GOEDEL-PROVER-V2-32B-SFT, Agentic SFT Training Data Volume=02026.04 | 4.25 | 9 | 3.5 | 1.5 | 4.56 | 6.14 | 5.35 | 95.83 |