Benchmark Almanac

M-RewardBench

GPT-4 Turbo Apr 2024 83.5% leads since October 2024
Not in model cards

Reward models choose preferred responses across chat, safety, reasoning and translation tasks in 23 languages. Accuracy measures preference selection.

The project evaluates reward models on 2,870 preference instances across 23 languages. The dataset has 37 Hugging Face likes and the repository has 45 GitHub stars.

Released
October 2024
Kind
Overall chat quality
Level
University
Scored
Checked automatically
Answers
Multiple choice
Tasks
All public
Measures
Preference
Language
Many languages
In and out
Text → Text
Interaction
Single turn
Items from
Another benchmark
Refresh
One fixed set
Run by
Anyone
Based on
RewardBench
In views
Many languages
Strongest model tested
GPT-4o May 2024 #162 on the capability index
Last model added
October 2024
Scores refreshed
November 21st, 2026
23
Entries on its own leaderboard, 9 models we track

GPT-4 Turbo leads with 83.5% among models released by October 2024 GPT-4 Turbo leads with 83.5% among models released by October 2024

Best score per entry from the benchmark's own leaderboard, read on November 21st, 2026. 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. This view shows only models we track, each at its best.

The newest model with a score here came out in October 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 GPT-4 Turbo Apr 2024, Generative RM 83.5%
  2. 2 GPT-4o May 2024, Generative RM 81.1%
  3. 3 google/gemma-2-9b-it Generative RM 76.5%
  4. 4 LxzGordon/URM-LLaMa-3.1-8B Sequence Classifier 76.3%
  5. 5 meta-llama/Meta-Llama-3.1-70B-Instruct Generative RM 75.5%
  6. 6 Aya Expanse 32B Generative RM 71.9%
  7. 7 meta-llama/Meta-Llama-3-70B-Instruct Generative RM 71.8%
  8. 8 CIR-AMS/BTRM_Qwen2_7b_0613 Sequence Classifier 70.6%
  9. 9 cohere/command-r-plus-08-2024 Generative RM 68.8%
  10. 10 allenai/tulu-2-dpo-13b DPO 68.2%
  11. 11 cohere/c4ai-aya-23-35b Generative RM 67.7%
  12. 12 Eurus-RM-7b Sequence Classifier 67.3%
  13. 13 NousResearch/Nous-Hermes-2-Mistral-7B-DPO DPO 67.3%
  14. 14 allenai/tulu-v2.5-13b-preference-mix-rm Sequence Classifier 66.8%
  15. 15 Zephyr 7B Beta DPO 65.7%
  16. 16 Aya Expanse 8B Generative RM 65.2%
  17. 17 meta-llama/Meta-Llama-3.1-8B-Instruct Generative RM 63.8%
  18. 18 c4ai-command-r-08-2024 Generative RM 63.5%
  19. 19 meta-llama/Meta-Llama-3-8B-Instruct Generative RM 62.8%
  20. 20 mistralai/Mistral-7B-Instruct-v0.3 Generative RM 61%
  21. 21 stabilityai/stablelm-zephyr-3b DPO 60.5%
  22. 22 Mistral-7B-Instruct-v0.2 Generative RM 59.8%
  23. 23 Gemma 1.1 7B IT Generative RM 58.4%

Source: Official M-RewardBench leaderboard

#EntryScoreSettingDateSource
1 GPT-4 Turbo Apr 2024 83.5% Generative RM Apr 2024 Official leaderboard
2 GPT-4o May 2024 81.1% Generative RM May 2024 Official leaderboard
3 google/gemma-2-9b-it 76.5% Generative RM Official leaderboard
4 LxzGordon/URM-LLaMa-3.1-8B 76.3% Sequence Classifier Official leaderboard
5 meta-llama/Meta-Llama-3.1-70B-Instruct 75.5% Generative RM Official leaderboard
6 Aya Expanse 32B 71.9% Generative RM Oct 2024 Official leaderboard
7 meta-llama/Meta-Llama-3-70B-Instruct 71.8% Generative RM Official leaderboard
8 CIR-AMS/BTRM_Qwen2_7b_0613 70.6% Sequence Classifier Official leaderboard
9 cohere/command-r-plus-08-2024 68.8% Generative RM Official leaderboard
10 allenai/tulu-2-dpo-13b 68.2% DPO Official leaderboard
11 cohere/c4ai-aya-23-35b 67.7% Generative RM Official leaderboard
12 Eurus-RM-7b 67.3% Sequence Classifier Apr 2024 Official leaderboard
13 NousResearch/Nous-Hermes-2-Mistral-7B-DPO 67.3% DPO Official leaderboard
14 allenai/tulu-v2.5-13b-preference-mix-rm 66.8% Sequence Classifier Official leaderboard
15 Zephyr 7B Beta 65.7% DPO Oct 2023 Official leaderboard
16 Aya Expanse 8B 65.2% Generative RM Oct 2024 Official leaderboard
17 meta-llama/Meta-Llama-3.1-8B-Instruct 63.8% Generative RM Official leaderboard
18 c4ai-command-r-08-2024 63.5% Generative RM Aug 2024 Official leaderboard
19 meta-llama/Meta-Llama-3-8B-Instruct 62.8% Generative RM Official leaderboard
20 mistralai/Mistral-7B-Instruct-v0.3 61% Generative RM Official leaderboard
21 stabilityai/stablelm-zephyr-3b 60.5% DPO Official leaderboard
22 Mistral-7B-Instruct-v0.2 59.8% Generative RM Dec 2023 Official leaderboard
23 Gemma 1.1 7B IT 58.4% Generative RM Apr 2024 Official leaderboard
#ModelScoreSettingDateSource
1 GPT-4 Turbo Apr 2024 83.5% Generative RM Apr 2024 Official leaderboard
2 GPT-4o May 2024 81.1% Generative RM May 2024 Official leaderboard
3 Aya Expanse 32B 71.9% Generative RM Oct 2024 Official leaderboard
4 Eurus-RM-7b 67.3% Sequence Classifier Apr 2024 Official leaderboard
5 Zephyr 7B Beta 65.7% DPO Oct 2023 Official leaderboard
6 Aya Expanse 8B 65.2% Generative RM Oct 2024 Official leaderboard
7 c4ai-command-r-08-2024 63.5% Generative RM Aug 2024 Official leaderboard
8 Mistral-7B-Instruct-v0.2 59.8% Generative RM Dec 2023 Official leaderboard
9 Gemma 1.1 7B IT 58.4% Generative RM Apr 2024 Official leaderboard

Rankings by part

We read 24 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as English, Arabic and Turkish. GPT-4 Turbo leads every one of them. Arrows show how far a model moved from its overall rank among the models we track.

English

9 models we track have a score here.

  1. 1 GPT-4 Turbo Apr 2024 85.4%
  2. 2 GPT-4o May 2024 85%
  3. 3 Eurus-RM-7b ↑1 80.4%
  4. 4 Zephyr 7B Beta ↑1 75.5%
  5. 5 Aya Expanse 32B ↓2 74.5%
  6. 6 Aya Expanse 8B 68.2%
  7. 7 c4ai-command-r-08-2024 67.2%
  8. 8 Mistral-7B-Instruct-v0.2 65.5%
  9. 9 Gemma 1.1 7B IT 60.5%

Number one over time

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

OpenAI
Own leaderboard Not tested since Oct 2024
GPT-4 Turbo 83.5%

Source: Official M-RewardBench 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 2025 Near the ceiling (85%) GPT-4 Turbo Eurus-RM-7b Zephyr 7B Beta 0% 20% 40% 60% 80% 100% 2024 2025 Near the ceiling (85%) GPT-4 Turbo Eurus-RM-7b

Source: Official M-RewardBench 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 October 2024, for Aya Expanse 32B. 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 (23 entries).
  • Pass Models scored 9 models we track, plus 14 other entries on its leaderboard.
  • Warning Newest score The newest came 2 years ago, in October 2024, for Aya Expanse 32B.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-4o (#162).
  • Pass Headroom The best is 83.5%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on huggingface.co: 9 rows for models we track and 14 other entries at the last read.
  • Pass Last read November 21st, 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