Preference Proxy Evaluations
Reward models and judges rank responses. The benchmark compares their decisions against user preferences and verifiable answers.
PPE reports separate results for human preferences and for correctness across several source benchmarks.
- Made by
- Arena
- Released
- October 2024
- Kind
- Overall chat quality
- Level
- University
- Scored
- Checked automatically
- Answers
- Short answer
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Items from
- Another benchmark
- Refresh
- One fixed set
- Run by
- Anyone
- Strongest model tested
- Claude 3.5 Sonnet #158 on the capability index
- Last model added
- September 2024
- Scores refreshed
- November 20th, 2026
Athene-RM-70B tops the leaderboard with 68.7% Claude 3.5 Sonnet leads with 68.5% among models released by September 2024
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Plain mean of the six dataset-level accuracies. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is Claude 3.5 Sonnet, at 68.5%. This view shows only models we track, each at its best.
The newest model with a score here came out in September 2024, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 68.7 Athene-RM-70B
- 68.5 Claude 3.5 Sonnet
- 65.0 llama-3.1-70b-instruct
- 64.1 GPT-4o mini
- 63.4 Llama-3-OffsetBias-RM-8B
- 62.8 Athene-RM-8B
- 62.7
internlm2-20b-reward
- 61.2 ArmoRM-Llama3-8B-v0.1
- 61.0
Skywork-Reward-Llama-3.1-8B
- 60.8
Starling-RM-34B
- 60.6
internlm2-7b-reward
- 60.4 nemotron-4-340b-reward
- 60.4
Eurus-RM-7b
- 56.5
Skywork-Reward-Gemma-2-27B
- 54.2
internlm2-1_8b-reward
- 53.1 Starling-RM-7B-alpha
- 1 Athene-RM-70B 68.7%
- 2 Claude 3.5 Sonnet 68.5%
- 3 llama-3.1-70b-instruct 65%
- 4 GPT-4o mini 64.1%
- 5 Llama-3-OffsetBias-RM-8B 63.4%
- 6 Athene-RM-8B 62.8%
- 7
internlm2-20b-reward 62.7%
- 8 ArmoRM-Llama3-8B-v0.1 61.2%
- 9
Skywork-Reward-Llama-3.1-8B 61%
- 10
Starling-RM-34B 60.8%
- 11
internlm2-7b-reward 60.6%
- 12 nemotron-4-340b-reward 60.4%
- 13
Eurus-RM-7b 60.4%
- 14
Skywork-Reward-Gemma-2-27B 56.5%
- 15
internlm2-1_8b-reward 54.2%
- 16 Starling-RM-7B-alpha 53.1%
Source: Official Preference Proxy Evaluations leaderboard
- 1 Claude 3.5 Sonnet 68.5%
- 2 GPT-4o mini 64.1%
- 3
internlm2-20b-reward 62.7%
- 4
Skywork-Reward-Llama-3.1-8B 61%
- 5
Starling-RM-34B 60.8%
- 6
internlm2-7b-reward 60.6%
- 7
Eurus-RM-7b 60.4%
- 8
Skywork-Reward-Gemma-2-27B 56.5%
- 9
internlm2-1_8b-reward 54.2%
Source: Official Preference Proxy Evaluations leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Athene-RM-70B | 68.7% | Official leaderboard | ||
| 2 | Claude 3.5 Sonnet | 68.5% | arena-hard | Jun 2024 | Official leaderboard |
| 3 | llama-3.1-70b-instruct | 65% | arena-hard | Official leaderboard | |
| 4 | GPT-4o mini | 64.1% | arena-hard | Jul 2024 | Official leaderboard |
| 5 | Llama-3-OffsetBias-RM-8B | 63.4% | Official leaderboard | ||
| 6 | Athene-RM-8B | 62.8% | Official leaderboard | ||
| 7 | | 62.7% | Jun 2024 | Official leaderboard | |
| 8 | ArmoRM-Llama3-8B-v0.1 | 61.2% | Official leaderboard | ||
| 9 | | 61% | Sep 2024 | Official leaderboard | |
| 10 | | 60.8% | Feb 2024 | Official leaderboard | |
| 11 | | 60.6% | Jun 2024 | Official leaderboard | |
| 12 | nemotron-4-340b-reward | 60.4% | Official leaderboard | ||
| 13 | | 60.4% | Apr 2024 | Official leaderboard | |
| 14 | | 56.5% | Sep 2024 | Official leaderboard | |
| 15 | | 54.2% | Jun 2024 | Official leaderboard | |
| 16 | Starling-RM-7B-alpha | 53.1% | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude 3.5 Sonnet | 68.5% | arena-hard | Jun 2024 | Official leaderboard |
| 2 | GPT-4o mini | 64.1% | arena-hard | Jul 2024 | Official leaderboard |
| 3 | | 62.7% | Jun 2024 | Official leaderboard | |
| 4 | | 61% | Sep 2024 | Official leaderboard | |
| 5 | | 60.8% | Feb 2024 | Official leaderboard | |
| 6 | | 60.6% | Jun 2024 | Official leaderboard | |
| 7 | | 60.4% | Apr 2024 | Official leaderboard | |
| 8 | | 56.5% | Sep 2024 | Official leaderboard | |
| 9 | | 54.2% | Jun 2024 | Official leaderboard |
Rankings by part
We read 37 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Human Preference, MMLU-Pro and MATH. Claude 3.5 Sonnet leads 24 of them, and 5 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Human Preference
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 67.3%
- 2 GPT-4o mini 66.5%
- 3
Starling-RM-34B ↑2 62.9%
- 4
Skywork-Reward-Llama-3.1-8B 62.4%
- 5
internlm2-7b-reward ↑1 62%
- 6
Eurus-RM-7b ↑1 62%
- 7
internlm2-20b-reward ↓4 61%
- 8
internlm2-1_8b-reward ↑1 57.2%
- 9
Skywork-Reward-Gemma-2-27B ↓1 56.6%
MMLU-Pro
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 81.4%
- 2 GPT-4o mini 70.6%
- 3
Starling-RM-34B ↑2 67.7%
- 4
internlm2-20b-reward ↓1 67.5%
- 5
internlm2-7b-reward ↑1 66.7%
- 6
Skywork-Reward-Llama-3.1-8B ↓2 64.3%
- 7
Eurus-RM-7b 63.4%
- 8
internlm2-1_8b-reward ↑1 57.5%
- 9
Skywork-Reward-Gemma-2-27B ↓1 53.9%
MATH
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 86.3%
- 2 GPT-4o mini 80.8%
- 3
internlm2-7b-reward ↑3 72.6%
- 4
internlm2-20b-reward ↓1 70.3%
- 5
Skywork-Reward-Llama-3.1-8B ↓1 69.6%
- 6
Eurus-RM-7b ↑1 69.3%
- 7
Starling-RM-34B ↓2 66.4%
- 8
Skywork-Reward-Gemma-2-27B 62.7%
- 9
internlm2-1_8b-reward 58.4%
GPQA
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 63.2%
- 2
internlm2-20b-reward ↑1 57.5%
- 3 GPT-4o mini ↓1 57.1%
- 4
Starling-RM-34B ↑1 57%
- 5
Skywork-Reward-Llama-3.1-8B ↓1 56.5%
- 6
internlm2-7b-reward 54.6%
- 7
Eurus-RM-7b 53.9%
- 8
internlm2-1_8b-reward ↑1 53.7%
- 9
Skywork-Reward-Gemma-2-27B ↓1 52.7%
MBPP+
9 models we track have a score here.
- 1
Skywork-Reward-Gemma-2-27B ↑7 59.3%
- 2
internlm2-20b-reward ↑1 57.6%
- 3
Starling-RM-34B ↑2 54.6%
- 4
Eurus-RM-7b ↑3 54%
- 5 Claude 3.5 Sonnet ↓4 54%
- 6 GPT-4o mini ↓4 53.6%
- 7
Skywork-Reward-Llama-3.1-8B ↓3 51.6%
- 8
internlm2-1_8b-reward ↑1 44.6%
- 9
internlm2-7b-reward ↓3 43.9%
IFEval
9 models we track have a score here.
- 1
internlm2-7b-reward ↑5 64%
- 2
internlm2-20b-reward ↑1 62.5%
- 3
Skywork-Reward-Llama-3.1-8B ↑1 61.5%
- 4
Eurus-RM-7b ↑3 59.4%
- 5 Claude 3.5 Sonnet ↓4 58.7%
- 6
Starling-RM-34B ↓1 56.1%
- 7 GPT-4o mini ↓5 56%
- 8
internlm2-1_8b-reward ↑1 53.9%
- 9
Skywork-Reward-Gemma-2-27B ↓1 53.8%
Human Preference: chinese_prompt
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 67%
- 2 GPT-4o mini 65.3%
- 3
internlm2-20b-reward 64.6%
- 4
internlm2-7b-reward ↑2 63.9%
- 5
Starling-RM-34B 62.5%
- 6
Eurus-RM-7b ↑1 61.2%
- 7
internlm2-1_8b-reward ↑2 60.1%
- 8
Skywork-Reward-Gemma-2-27B 59.2%
- 9
Skywork-Reward-Llama-3.1-8B ↓5 57.2%
Human Preference: easy_prompt
9 models we track have a score here.
- 1 GPT-4o mini ↑1 67.5%
- 2 Claude 3.5 Sonnet ↓1 67%
- 3
Skywork-Reward-Llama-3.1-8B ↑1 63.2%
- 4
Eurus-RM-7b ↑3 61.8%
- 5
internlm2-7b-reward ↑1 61.7%
- 6
Starling-RM-34B ↓1 61.4%
- 7
internlm2-20b-reward ↓4 60.4%
- 8
Skywork-Reward-Gemma-2-27B 56.7%
- 9
internlm2-1_8b-reward 55.5%
Human Preference: english_prompt
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 66.4%
- 2 GPT-4o mini 66.4%
- 3
Starling-RM-34B ↑2 63.5%
- 4
Skywork-Reward-Llama-3.1-8B 62.9%
- 5
Eurus-RM-7b ↑2 61.8%
- 6
internlm2-7b-reward 61%
- 7
internlm2-20b-reward ↓4 59.3%
- 8
internlm2-1_8b-reward ↑1 56.2%
- 9
Skywork-Reward-Gemma-2-27B ↓1 55.2%
Human Preference: hard_prompt
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 69.3%
- 2 GPT-4o mini 66.6%
- 3
Starling-RM-34B ↑2 63.9%
- 4
Eurus-RM-7b ↑3 62.7%
- 5
internlm2-7b-reward ↑1 62.1%
- 6
internlm2-20b-reward ↓3 61.6%
- 7
Skywork-Reward-Llama-3.1-8B ↓3 61.1%
- 8
internlm2-1_8b-reward ↑1 58.5%
- 9
Skywork-Reward-Gemma-2-27B ↓1 56.2%
Human Preference: if_prompt
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 68.4%
- 2 GPT-4o mini 65.8%
- 3
internlm2-20b-reward 61.9%
- 4
internlm2-7b-reward ↑2 61.4%
- 5
Skywork-Reward-Llama-3.1-8B ↓1 61.4%
- 6
Starling-RM-34B ↓1 61.1%
- 7
Eurus-RM-7b 60.9%
- 8
internlm2-1_8b-reward ↑1 57.3%
- 9
Skywork-Reward-Gemma-2-27B ↓1 56.4%
Human Preference: is_code
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 68.1%
- 2 GPT-4o mini 66.1%
- 3
Starling-RM-34B ↑2 64.3%
- 4
internlm2-7b-reward ↑2 63.5%
- 5
Eurus-RM-7b ↑2 63%
- 6
internlm2-20b-reward ↓3 62.1%
- 7
Skywork-Reward-Llama-3.1-8B ↓3 61%
- 8
internlm2-1_8b-reward ↑1 58.8%
- 9
Skywork-Reward-Gemma-2-27B ↓1 56.4%
Human Preference: math_prompt
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 71.8%
- 2 GPT-4o mini 70.2%
- 3
Starling-RM-34B ↑2 63.9%
- 4
internlm2-20b-reward ↓1 62.6%
- 5
internlm2-7b-reward ↑1 61.5%
- 6
Eurus-RM-7b ↑1 61.3%
- 7
Skywork-Reward-Llama-3.1-8B ↓3 60.6%
- 8
internlm2-1_8b-reward ↑1 58.7%
- 9
Skywork-Reward-Gemma-2-27B ↓1 55.8%
Human Preference: non_english_prompt
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 68.5%
- 2 GPT-4o mini 66.5%
- 3
internlm2-7b-reward ↑3 63.4%
- 4
internlm2-20b-reward ↓1 63.1%
- 5
Eurus-RM-7b ↑2 62.3%
- 6
Starling-RM-34B ↓1 62.2%
- 7
Skywork-Reward-Llama-3.1-8B ↓3 61.7%
- 8
internlm2-1_8b-reward ↑1 58.6%
- 9
Skywork-Reward-Gemma-2-27B ↓1 58.4%
Human Preference: russian_prompt
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 68.6%
- 2 GPT-4o mini 65.5%
- 3
Eurus-RM-7b ↑4 62.8%
- 4
Skywork-Reward-Llama-3.1-8B 61.2%
- 5
internlm2-7b-reward ↑1 61.1%
- 6
Starling-RM-34B ↓1 61%
- 7
internlm2-20b-reward ↓4 60.4%
- 8
internlm2-1_8b-reward ↑1 57.6%
- 9
Skywork-Reward-Gemma-2-27B ↓1 56.3%
Human Preference: shorter_won
9 models we track have a score here.
- 1
Skywork-Reward-Llama-3.1-8B ↑3 54.7%
- 2
Skywork-Reward-Gemma-2-27B ↑6 54.5%
- 3
internlm2-20b-reward 52.4%
- 4 Claude 3.5 Sonnet ↓3 52.3%
- 5 GPT-4o mini ↓3 50.1%
- 6
internlm2-1_8b-reward ↑3 47.9%
- 7
internlm2-7b-reward ↓1 45.8%
- 8
Eurus-RM-7b ↓1 39.8%
- 9
Starling-RM-34B ↓4 39.2%
Human Preference: similar_response
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 67.1%
- 2 GPT-4o mini 66.3%
- 3
Starling-RM-34B ↑2 64.1%
- 4
Skywork-Reward-Llama-3.1-8B 62.8%
- 5
Eurus-RM-7b ↑2 62.1%
- 6
internlm2-7b-reward 60.8%
- 7
internlm2-20b-reward ↓4 59%
- 8
Skywork-Reward-Gemma-2-27B 55.9%
- 9
internlm2-1_8b-reward 55.9%
MMLU-Pro: Meta-Llama-3-8B-Instruct
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 82.8%
- 2
internlm2-20b-reward ↑1 72.7%
- 3 GPT-4o mini ↓1 72.5%
- 4
Starling-RM-34B ↑1 72.2%
- 5
internlm2-7b-reward ↑1 70.8%
- 6
Eurus-RM-7b ↑1 65.8%
- 7
Skywork-Reward-Llama-3.1-8B ↓3 64.7%
- 8
internlm2-1_8b-reward ↑1 58.1%
- 9
Skywork-Reward-Gemma-2-27B ↓1 56.9%
MMLU-Pro: claude-3-haiku-20240307
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 84.5%
- 2 GPT-4o mini 77%
- 3
internlm2-20b-reward 70.6%
- 4
internlm2-7b-reward ↑2 68.9%
- 5
Starling-RM-34B 68.1%
- 6
Skywork-Reward-Llama-3.1-8B ↓2 67.3%
- 7
Eurus-RM-7b 63%
- 8
internlm2-1_8b-reward ↑1 55.9%
- 9
Skywork-Reward-Gemma-2-27B ↓1 53.9%
MMLU-Pro: gemma-2-9b-it
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 80.5%
- 2 GPT-4o mini 72.1%
- 3
Skywork-Reward-Llama-3.1-8B ↑1 66.3%
- 4
Starling-RM-34B ↑1 65.8%
- 5
internlm2-7b-reward ↑1 65.8%
- 6
Eurus-RM-7b ↑1 65.5%
- 7
internlm2-20b-reward ↓4 60.2%
- 8
internlm2-1_8b-reward ↑1 59.7%
- 9
Skywork-Reward-Gemma-2-27B ↓1 57.3%
MMLU-Pro: gpt-4o-mini-2024-07-18
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 77.7%
- 2
internlm2-20b-reward ↑1 66.7%
- 3
Starling-RM-34B ↑2 64.5%
- 4
internlm2-7b-reward ↑2 61.4%
- 5 GPT-4o mini ↓3 60.8%
- 6
Eurus-RM-7b ↑1 59.5%
- 7
Skywork-Reward-Llama-3.1-8B ↓3 58.9%
- 8
internlm2-1_8b-reward ↑1 56.4%
- 9
Skywork-Reward-Gemma-2-27B ↓1 47.5%
MATH: Meta-Llama-3-8B-Instruct
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 90.5%
- 2 GPT-4o mini 87.9%
- 3
internlm2-7b-reward ↑3 74.1%
- 4
internlm2-20b-reward ↓1 72.3%
- 5
Eurus-RM-7b ↑2 70.9%
- 6
Skywork-Reward-Llama-3.1-8B ↓2 67.5%
- 7
Starling-RM-34B ↓2 66.7%
- 8
Skywork-Reward-Gemma-2-27B 65.2%
- 9
internlm2-1_8b-reward 61.1%
MATH: claude-3-haiku-20240307
9 models we track have a score here.
- 1 GPT-4o mini ↑1 84%
- 2 Claude 3.5 Sonnet ↓1 83.4%
- 3
internlm2-7b-reward ↑3 72.7%
- 4
Eurus-RM-7b ↑3 68.4%
- 5
Skywork-Reward-Llama-3.1-8B ↓1 67.7%
- 6
internlm2-20b-reward ↓3 66.1%
- 7
Starling-RM-34B ↓2 61.6%
- 8
Skywork-Reward-Gemma-2-27B 58.1%
- 9
internlm2-1_8b-reward 58%
MATH: gemma-2-9b-it
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 88.8%
- 2 GPT-4o mini 87.5%
- 3
internlm2-7b-reward ↑3 80.9%
- 4
internlm2-20b-reward ↓1 75.8%
- 5
Skywork-Reward-Llama-3.1-8B ↓1 75.2%
- 6
Eurus-RM-7b ↑1 73.8%
- 7
Starling-RM-34B ↓2 71.7%
- 8
Skywork-Reward-Gemma-2-27B 68.9%
- 9
internlm2-1_8b-reward 61.6%
MATH: gpt-4o-mini-2024-07-18
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 82.5%
- 2
Skywork-Reward-Llama-3.1-8B ↑2 68.3%
- 3
internlm2-20b-reward 67%
- 4
Starling-RM-34B ↑1 65.5%
- 5
Eurus-RM-7b ↑2 63.9%
- 6 GPT-4o mini ↓4 63.9%
- 7
internlm2-7b-reward ↓1 62.7%
- 8
Skywork-Reward-Gemma-2-27B 58.6%
- 9
internlm2-1_8b-reward 53.1%
GPQA: Meta-Llama-3-8B-Instruct
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 60.9%
- 2 GPT-4o mini 59.1%
- 3
Starling-RM-34B ↑2 58.8%
- 4
internlm2-20b-reward ↓1 56.4%
- 5
Skywork-Reward-Gemma-2-27B ↑3 55.3%
- 6
Eurus-RM-7b ↑1 55.2%
- 7
internlm2-1_8b-reward ↑2 54.8%
- 8
internlm2-7b-reward ↓2 54.1%
- 9
Skywork-Reward-Llama-3.1-8B ↓5 50.9%
GPQA: claude-3-haiku-20240307
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 60.9%
- 2
Skywork-Reward-Llama-3.1-8B ↑2 55.9%
- 3
internlm2-7b-reward ↑3 55.8%
- 4 GPT-4o mini ↓2 54.7%
- 5
Eurus-RM-7b ↑2 53.9%
- 6
internlm2-20b-reward ↓3 52.3%
- 7
Starling-RM-34B ↓2 51.6%
- 8
Skywork-Reward-Gemma-2-27B 51.1%
- 9
internlm2-1_8b-reward 50.8%
GPQA: gemma-2-9b-it
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 64.7%
- 2
internlm2-20b-reward ↑1 60.5%
- 3
Skywork-Reward-Llama-3.1-8B ↑1 60%
- 4 GPT-4o mini ↓2 58.3%
- 5
internlm2-7b-reward ↑1 58%
- 6
Starling-RM-34B ↓1 57.5%
- 7
internlm2-1_8b-reward ↑2 57.3%
- 8
Skywork-Reward-Gemma-2-27B 56.1%
- 9
Eurus-RM-7b ↓2 53.8%
GPQA: gpt-4o-mini-2024-07-18
9 models we track have a score here.
- 1 Claude 3.5 Sonnet 66.4%
- 2
internlm2-20b-reward ↑1 60.6%
- 3
Starling-RM-34B ↑2 60.2%
- 4
Skywork-Reward-Llama-3.1-8B 59.2%
- 5 GPT-4o mini ↓3 56.6%
- 6
Eurus-RM-7b ↑1 53%
- 7
internlm2-1_8b-reward ↑2 51.7%
- 8
internlm2-7b-reward ↓2 50.6%
- 9
Skywork-Reward-Gemma-2-27B ↓1 48.4%
MBPP+: Meta-Llama-3-8B-Instruct
9 models we track have a score here.
- 1 GPT-4o mini ↑1 76.3%
- 2 Claude 3.5 Sonnet ↓1 74.3%
- 3
Eurus-RM-7b ↑4 71%
- 4
Skywork-Reward-Llama-3.1-8B 71%
- 5
internlm2-7b-reward ↑1 70.1%
- 6
internlm2-20b-reward ↓3 68.9%
- 7
internlm2-1_8b-reward ↑2 63.3%
- 8
Starling-RM-34B ↓3 62.5%
- 9
Skywork-Reward-Gemma-2-27B ↓1 59.6%
MBPP+: claude-3-haiku-20240307
9 models we track have a score here.
- 1
Skywork-Reward-Gemma-2-27B ↑7 61.5%
- 2
internlm2-20b-reward ↑1 49.2%
- 3
Starling-RM-34B ↑2 48.8%
- 4 Claude 3.5 Sonnet ↓3 44.3%
- 5
Eurus-RM-7b ↑2 41.5%
- 6 GPT-4o mini ↓4 41.4%
- 7
Skywork-Reward-Llama-3.1-8B ↓3 36.5%
- 8
internlm2-1_8b-reward ↑1 28.8%
- 9
internlm2-7b-reward ↓3 22.1%
MBPP+: gemma-2-9b-it
9 models we track have a score here.
- 1 GPT-4o mini ↑1 74.2%
- 2 Claude 3.5 Sonnet ↓1 69.5%
- 3
Skywork-Reward-Llama-3.1-8B ↑1 68.4%
- 4
Eurus-RM-7b ↑3 65.8%
- 5
internlm2-20b-reward ↓2 65.8%
- 6
internlm2-7b-reward 61.6%
- 7
internlm2-1_8b-reward ↑2 60%
- 8
Starling-RM-34B ↓3 59.5%
- 9
Skywork-Reward-Gemma-2-27B ↓1 51.6%
MBPP+: gpt-4o-mini-2024-07-18
9 models we track have a score here.
- 1
internlm2-20b-reward ↑2 57.5%
- 2
Starling-RM-34B ↑3 54.6%
- 3
Skywork-Reward-Gemma-2-27B ↑5 54.3%
- 4
Eurus-RM-7b ↑3 54%
- 5
internlm2-7b-reward ↑1 53.3%
- 6
Skywork-Reward-Llama-3.1-8B ↓2 52.4%
- 7 Claude 3.5 Sonnet ↓6 51.1%
- 8
internlm2-1_8b-reward ↑1 50.8%
- 9 GPT-4o mini ↓7 50.5%
IFEval: Meta-Llama-3-8B-Instruct
9 models we track have a score here.
- 1
internlm2-7b-reward ↑5 69.2%
- 2
internlm2-20b-reward ↑1 65.8%
- 3
Eurus-RM-7b ↑4 62.3%
- 4 Claude 3.5 Sonnet ↓3 61%
- 5
Skywork-Reward-Llama-3.1-8B ↓1 59.5%
- 6 GPT-4o mini ↓4 57.8%
- 7
Starling-RM-34B ↓2 57.8%
- 8
internlm2-1_8b-reward ↑1 57.5%
- 9
Skywork-Reward-Gemma-2-27B ↓1 56.4%
IFEval: claude-3-haiku-20240307
9 models we track have a score here.
- 1
Skywork-Reward-Llama-3.1-8B ↑3 64.1%
- 2
internlm2-7b-reward ↑4 58.9%
- 3 Claude 3.5 Sonnet ↓2 57.2%
- 4
internlm2-20b-reward ↓1 57%
- 5 GPT-4o mini ↓3 55.9%
- 6
Starling-RM-34B ↓1 51.4%
- 7
internlm2-1_8b-reward ↑2 51.3%
- 8
Skywork-Reward-Gemma-2-27B 50.9%
- 9
Eurus-RM-7b ↓2 50.9%
IFEval: gemma-2-9b-it
9 models we track have a score here.
- 1
internlm2-20b-reward ↑2 64.1%
- 2
Eurus-RM-7b ↑5 63.9%
- 3
internlm2-7b-reward ↑3 63%
- 4
Skywork-Reward-Llama-3.1-8B 60.8%
- 5 Claude 3.5 Sonnet ↓4 60.1%
- 6 GPT-4o mini ↓4 57.5%
- 7
internlm2-1_8b-reward ↑2 55.2%
- 8
Skywork-Reward-Gemma-2-27B 55.2%
- 9
Starling-RM-34B ↓4 54.7%
IFEval: gpt-4o-mini-2024-07-18
9 models we track have a score here.
- 1
internlm2-7b-reward ↑5 65%
- 2
internlm2-20b-reward ↑1 63%
- 3
Skywork-Reward-Llama-3.1-8B ↑1 61.6%
- 4
Starling-RM-34B ↑1 60.6%
- 5
Eurus-RM-7b ↑2 60.3%
- 6 Claude 3.5 Sonnet ↓5 56.4%
- 7 GPT-4o mini ↓5 54%
- 8
Skywork-Reward-Gemma-2-27B 52.7%
- 9
internlm2-1_8b-reward 51.6%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official Preference Proxy Evaluations leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official Preference Proxy Evaluations leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Instructions and chat
- IFEval In use Models answer about 500 prompts containing verifiable instructions such as word limits or required keywords. Code checks each instruction for strict and loose accuracy.
Gemma 4 31B IT 98.9%
- IFBench In use 58 new out-of-domain verifiable output constraints. Scored by programmatic constraint checks (strict or loose). Flint 89.2%
- MultiChallenge In use Multi-turn conversations testing instruction retention, inference memory, versioned editing and self-coherence. Graded by LLM judge with rubrics. GPT-6.1 Sol 82%
- Multi-IF In use Models follow verifiable instructions over three-turn conversations in eight languages, extending IFEval. Scored by instruction accuracy at each turn. LFM3-1.7B 89.1%
- AdvancedIF In use Models respond to over 1,600 prompts with complex, multi-turn, and system-level instructions. A judge model checks each response against expert-written rubrics.
MAI-Thinking-1 85%
- MM-IFEval In use Multimodal models follow image-grounded instructions with exact output constraints. Checks mix rule-based verification and a judge model. GPT-4o 64.6%
Data audit
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 2 years ago, in September 2024, for Skywork-Reward-Gemma-2-27B. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
- Pass Independent runs Scores from the official leaderboard (16 entries).
- Pass Models scored 9 models we track, plus 7 other entries on its leaderboard.
- Warning Newest score The newest came 2 years ago, in September 2024, for Skywork-Reward-Gemma-2-27B.
- Note Frontier models None of the top 10 has a score. The strongest tested is Claude 3.5 Sonnet (#158).
- Pass Headroom The best is 68.5%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on huggingface.co: 11 rows for models we track and 7 other entries at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.