List your company
ResearchFormerly WizWand

Mathematical Reasoning on AIME 2024 (Pass@1, Pass@32, Avg)

57.29Pass@1FG-ExPO
-2.291613.176728.64544.1133Sep 28, 2025Nov 9, 2025Dec 21, 2025Feb 1, 2026Mar 15, 2026Apr 26, 2026Jun 7, 2026
Updated 4mo ago

Evaluation results

MethodLinks
FG-ExPOBackbone=Qwen3-8B-Base2026.0557.2980—
GRPOBackbone=Qwen3-8B-Base2026.0554.5880—
ISPOBase Model=Qwen2.5-Math-7B, Decoding Strategy=Greedy (T=0)2026.0648.9——
Qwen3-8B-BaseTraining=RLAD (ours), Context=30K2026.0247.885.466.5
PACRBase Model=Qwen2.5-Math-7B, Decoding Strategy=Greedy (T=0)2026.0643.3——
Scaf-GRPOBase Model=Qwen2.5-Math-7B, Decoding Strategy=Greedy (T=0)2026.0643.3——
Qwen3-8B-BaseTraining=KDRL, Context=30K2026.0242.186.164.9
CoVerRLBackbone=Qwen3-4B2026.0540.5——
SCOPEBackbone=Qwen3-4B2026.0538.8——
TTRL (w/ MPS)Backbone=Qwen3-4B2026.0538——
TTRL-GuardBackbone=Qwen3-4B2026.0537.8——
Qwen3-8B-BaseTraining=GRPO, Context=30K2026.0237.777.861
TTRLBackbone=Qwen3-4B2026.0536.7——
TTRL (w/ RCSU)Backbone=Qwen3-4B2026.0536.4——
TTRL (w/ FRS)Backbone=Qwen3-4B2026.0535.5——
GRPO-SBase Model=Qwen2.5-7B, Decoding Strategy=Greedy (T=0)2026.0634.7——
GTPOBase Model=Qwen2.5-7B, Decoding Strategy=Greedy (T=0)2026.0634.2——
ISPOBase Model=Qwen2.5-7B, Decoding Strategy=Greedy (T=0)2026.0632.2——
Vanilla GRPOBase Model=Qwen2.5-Math-7B, Decoding Strategy=Greedy (T=0)2026.0630——
Dr. GRPOBase Model=Qwen2.5-Math-7B, Decoding Strategy=Greedy (T=0)2026.0630——
FG-ExPOBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0529.0683.33—
ISPOBase Model=Qwen2.5-Math-1.5B, Decoding Strategy=Greedy (T=0)2026.0627.8——
GRPOBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0527.473.33—
Qwen3-8B-BaseTraining=SFT, Context=30K2026.0226.765.856
PREPOBase Model=Qwen2.5-Math-7B, Decoding Strategy=Greedy (T=0)2026.0626.2——
PACRBase Model=Qwen2.5-Math-1.5B, Decoding Strategy=Greedy (T=0)2026.0623.3——
TTRL-GuardBackbone=Qwen2.5-7B-Instruct2026.0521.9——
CoVerRLBackbone=Qwen2.5-7B-Instruct2026.0521.5——
TTRL (w/ RCSU)Backbone=Qwen2.5-7B-Instruct2026.0521.3——
ProGRPOBase Model=Qwen2.5-7B, Decoding Strategy=Greedy (T=0)2026.0621.3——
SCOPEBackbone=Qwen2.5-7B-Instruct2026.0521.1——
BaseBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0520.7363.33—
TTRL (w/ MPS)Backbone=Qwen2.5-7B-Instruct2026.0520.5——
TTRL (w/ MPS)Backbone=Llama-3.2-3B-Instruct2026.0520.3——
TTRL (w/ FRS)Backbone=Qwen2.5-7B-Instruct2026.0520.2——
TTRLBackbone=Qwen2.5-7B-Instruct2026.0520——
Scaf-GRPOBase Model=Qwen2.5-Math-1.5B, Decoding Strategy=Greedy (T=0)2026.0620——
TTRL-GuardBackbone=Llama-3.2-3B-Instruct2026.0519.5——
Base ModelBackbone=Qwen2.5-7B-Instruct2026.0518.1——
SCOPEBackbone=Llama-3.2-3B-Instruct2026.0517.1——
TTRL (w/ FRS)Backbone=Llama-3.2-3B-Instruct2026.0517.1——
CoVerRLBackbone=Llama-3.2-3B-Instruct2026.0516.9——
TTRL (w/ RCSU)Backbone=Llama-3.2-3B-Instruct2026.0516.7——
PREPOBase Model=Qwen2.5-Math-1.5B, Decoding Strategy=Greedy (T=0)2026.0616.7——
Base modelBase Model=Qwen2.5-Math-7B, Decoding Strategy=Greedy (T=0)2026.0616.7——
PREPOBase Model=Qwen2.5-7B, Decoding Strategy=Greedy (T=0)2026.0616.1——
Qwen3-8B-BaseTraining=-, Context=30K2026.0213.943.636.8
Llama-3.2-3B-Instruct + GRPO w/ VERL.Backbone=Llama-3.2-3B-Instruct, RL Method=GRPO, VERL=Enabled2025.0913.3——
Qwen2.5-7B + GRPO w/ VERL.Backbone=Qwen2.5-7B, RL Method=GRPO, VERL=Enabled2025.0913.3——
TTRLBackbone=Llama-3.2-3B-Instruct2026.0513.3——
Vanilla GRPOBase Model=Qwen2.5-Math-1.5B, Decoding Strategy=Greedy (T=0)2026.0613.3——
Dr. GRPOBase Model=Qwen2.5-Math-1.5B, Decoding Strategy=Greedy (T=0)2026.0613.3——
Scaf-GRPOBase Model=Qwen2.5-7B, Decoding Strategy=Greedy (T=0)2026.0613.3——
Qwen3-1.7B-BaseTraining=RLAD (ours), Context=30K2026.0212.135.139
Qwen3-1.7B-BaseTraining=KDRL, Context=30K2026.0210.232.837.2
Qwen3-1.7B-BaseTraining=GRPO, Context=30K2026.021030.836.5
Llama-3.2-3B-Instruct + PPOBackbone=Llama-3.2-3B-Instruct, RL Method=PPO2025.0910——
Llama-3.2-3B-Instruct + PPO w/ VERL.Backbone=Llama-3.2-3B-Instruct, RL Method=PPO, VERL=Enabled2025.0910——
Llama-3.1-8B-Instruct + GRPO w/ VERL.Backbone=Llama-3.1-8B-Instruct, RL Method=GRPO, VERL=Enabled2025.0910——
Llama-3.1-8B-Instruct + PPO w/ VERL.Backbone=Llama-3.1-8B-Instruct, RL Method=PPO, VERL=Enabled2025.0910——
Qwen2.5-7B + GRPOBackbone=Qwen2.5-7B, RL Method=GRPO2025.0910——
Qwen2.5-7B + PPO w/ VERL.Backbone=Qwen2.5-7B, RL Method=PPO, VERL=Enabled2025.0910——
Mathstral-7B-v0.1 + PPO w/ VERL.Backbone=Mathstral-7B-v0.1, RL Method=PPO, VERL=Enabled2025.0910——
Vanilla GRPOBase Model=Qwen2.5-7B, Decoding Strategy=Greedy (T=0)2026.069.2——
Base modelBase Model=Qwen2.5-Math-1.5B, Decoding Strategy=Greedy (T=0)2026.067.2——
Llama-3.1-8B-Instruct + GRPOBackbone=Llama-3.1-8B-Instruct, RL Method=GRPO2025.096.7——
Llama-3.1-8B-Instruct + PPOBackbone=Llama-3.1-8B-Instruct, RL Method=PPO2025.096.7——
Qwen2.5-3BBackbone=Qwen2.5-3B, RL Method=None2025.096.7——
Qwen2.5-3B + GRPO w/ VERL.Backbone=Qwen2.5-3B, RL Method=GRPO, VERL=Enabled2025.096.7——
Qwen2.5-3B + PPO w/ VERL.Backbone=Qwen2.5-3B, RL Method=PPO, VERL=Enabled2025.096.7——
Qwen2.5-7BBackbone=Qwen2.5-7B, RL Method=None2025.096.7——
Qwen2.5-7B + PPOBackbone=Qwen2.5-7B, RL Method=PPO2025.096.7——
Mathstral-7B-v0.1 + GRPO w/ VERL.Backbone=Mathstral-7B-v0.1, RL Method=GRPO, VERL=Enabled2025.096.7——
Mathstral-7B-v0.1 + PPOBackbone=Mathstral-7B-v0.1, RL Method=PPO2025.096.7——
Base modelBase Model=Qwen2.5-7B, Decoding Strategy=Greedy (T=0)2026.065.4——
Base ModelBackbone=Llama-3.2-3B-Instruct2026.054.7——
Llama-3.2-3B-Instruct + GRPOBackbone=Llama-3.2-3B-Instruct, RL Method=GRPO2025.093.3——
Qwen2.5-3B + GRPOBackbone=Qwen2.5-3B, RL Method=GRPO2025.093.3——
Qwen2.5-3B + PPOBackbone=Qwen2.5-3B, RL Method=PPO2025.093.3——
Base ModelBackbone=Qwen3-4B2026.053.3——
Qwen3-1.7B-BaseTraining=-, Context=30K2026.022.120.822.2
Llama-3.2-3B-InstructBackbone=Llama-3.2-3B-Instruct, RL Method=None2025.090——
Llama-3.1-8B-InstructBackbone=Llama-3.1-8B-Instruct, RL Method=None2025.090——
Mathstral-7B-v0.1Backbone=Mathstral-7B-v0.1, RL Method=None2025.090——
Mathstral-7B-v0.1 + GRPOBackbone=Mathstral-7B-v0.1, RL Method=GRPO2025.090——
Mistral-7B-v0.3Backbone=Mistral-7B-v0.3, RL Method=None2025.090——
Mistral-7B-v0.3 + GRPOBackbone=Mistral-7B-v0.3, RL Method=GRPO2025.090——
Mistral-7B-v0.3 + GRPO w/ VERL.Backbone=Mistral-7B-v0.3, RL Method=GRPO, VERL=Enabled2025.090——
Mistral-7B-v0.3 + PPOBackbone=Mistral-7B-v0.3, RL Method=PPO2025.090——
Mistral-7B-v0.3 + PPO w/ VERL.Backbone=Mistral-7B-v0.3, RL Method=PPO, VERL=Enabled2025.090——