Benchmark Almanac

Preference Proxy Evaluations

Athene-RM-70B 68.7% tops its leaderboard Claude 3.5 Sonnet 68.5% leads the models we track
Not in model cards

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
In and out
Text → Text
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
16
Entries on its own leaderboard, 9 models we track

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:

  1. 1 Athene-RM-70B 68.7%
  2. 2 Claude 3.5 Sonnet 68.5%
  3. 3 llama-3.1-70b-instruct 65%
  4. 4 GPT-4o mini 64.1%
  5. 5 Llama-3-OffsetBias-RM-8B 63.4%
  6. 6 Athene-RM-8B 62.8%
  7. 7 internlm2-20b-reward 62.7%
  8. 8 ArmoRM-Llama3-8B-v0.1 61.2%
  9. 9 Skywork-Reward-Llama-3.1-8B 61%
  10. 10 Starling-RM-34B 60.8%
  11. 11 internlm2-7b-reward 60.6%
  12. 12 nemotron-4-340b-reward 60.4%
  13. 13 Eurus-RM-7b 60.4%
  14. 14 Skywork-Reward-Gemma-2-27B 56.5%
  15. 15 internlm2-1_8b-reward 54.2%
  16. 16 Starling-RM-7B-alpha 53.1%

Source: Official Preference Proxy Evaluations leaderboard

#EntryScoreSettingDateSource
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 internlm2-20b-reward 62.7% Jun 2024 Official leaderboard
8 ArmoRM-Llama3-8B-v0.1 61.2% Official leaderboard
9 Skywork-Reward-Llama-3.1-8B 61% Sep 2024 Official leaderboard
10 Starling-RM-34B 60.8% Feb 2024 Official leaderboard
11 internlm2-7b-reward 60.6% Jun 2024 Official leaderboard
12 nemotron-4-340b-reward 60.4% Official leaderboard
13 Eurus-RM-7b 60.4% Apr 2024 Official leaderboard
14 Skywork-Reward-Gemma-2-27B 56.5% Sep 2024 Official leaderboard
15 internlm2-1_8b-reward 54.2% Jun 2024 Official leaderboard
16 Starling-RM-7B-alpha 53.1% Official leaderboard
#ModelScoreSettingDateSource
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 internlm2-20b-reward 62.7% Jun 2024 Official leaderboard
4 Skywork-Reward-Llama-3.1-8B 61% Sep 2024 Official leaderboard
5 Starling-RM-34B 60.8% Feb 2024 Official leaderboard
6 internlm2-7b-reward 60.6% Jun 2024 Official leaderboard
7 Eurus-RM-7b 60.4% Apr 2024 Official leaderboard
8 Skywork-Reward-Gemma-2-27B 56.5% Sep 2024 Official leaderboard
9 internlm2-1_8b-reward 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. 1 Claude 3.5 Sonnet 67.3%
  2. 2 GPT-4o mini 66.5%
  3. 3 Starling-RM-34B ↑2 62.9%
  4. 4 Skywork-Reward-Llama-3.1-8B 62.4%
  5. 5 internlm2-7b-reward ↑1 62%
  6. 6 Eurus-RM-7b ↑1 62%
  7. 7 internlm2-20b-reward ↓4 61%
  8. 8 internlm2-1_8b-reward ↑1 57.2%
  9. 9 Skywork-Reward-Gemma-2-27B ↓1 56.6%

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

Anthropic
Own leaderboard Not tested since Oct 2024

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.

Plot with other benchmarks
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2024 Near the ceiling (85%) Claude 3.5 Sonnet Starling-RM-34B 0% 20% 40% 60% 80% 100% 2024 Near the ceiling (85%) Claude 3.5 Sonnet Starling-RM-34B

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

All 144 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.

How we check