Our new X account is live! Follow @wizwand_team for updates

Multi-turn Conversation Rating (1-10) on MT-Bench

8.7Conversation Rating (1-10)

GRPO (RM)

Updated 4d ago

Evaluation Results

Method	Links
GRPO (RM) 2026.01		8.7
GRPO (RLVRR) 2026.01		8.7
DPO 2026.01		8.6
SFT 2026.01		8.5
SFT 2026.01		8.5
Instruct 2026.01		8.4
GRPO (GRM) 2026.01		8.4
GRPO (RLPR) 2026.01		8.3
GRPO (BLEU) 2026.01		8.2
vs. Qwen3-8B 2026.02		8.15
Elo-Evolve 2026.02		8.04
Elo-Evolve 2026.02		8.03
vs. Qwen2.5-14B 2026.02		7.99
vs. Qwen2.5-14B 2026.02		7.99
vs. Qwen2.5-14B 2026.02		7.98
DNO (replicated) 2026.02		7.97
vs. Qwen2.5-32B 2026.02		7.96
DNO (replicated) 2026.02		7.95
DNO (replicated) 2026.02		7.92
Point GRPO 2026.02		7.91
vs. Qwen2.5-32B 2026.02		7.89
vs. Qwen3-8B 2026.02		7.86
Qwen2.5-7B (base model) 2026.02		7.84
Elo-Evolve 2026.02		7.82
Point GRPO 2026.02		7.81
vs. Qwen3-8B 2026.02		7.81
Point GRPO 2026.02		7.79
vs. Qwen2.5-32B 2026.02		7.79
RevFFN 2025.12		7.65
GRPO (Random) 2026.01		7.6
SFT + Checkpointing 2025.12		7.52
LOMO 2025.12		7.5
GaLore 2025.12		7.46
DoRA 2025.12		7.25
LoRA 2025.12		7.18
(IA)³ 2025.12		7.15
ADPO 2026.02		7.1
DPO 2026.02		7
SimPO 2026.02		7
SFT 2026.02		6.9
Base Model 2025.12		6.25