M-RewardBench
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.
- Made by
- Cohere Labs Community
- Released
- October 2024
- Kind
- Overall chat quality
- Level
- University
- Scored
- Checked automatically
- Answers
- Multiple choice
- Tasks
- All public
- Measures
- Preference
- Language
- Many languages
- 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
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:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 83.5 GPT-4 Turbo
- 81.1 GPT-4o
- 76.5 google/gemma-2-9b-it
- 76.3 LxzGordon/URM-LLaMa-3.1-8B
- 75.5 meta-llama/Meta-Llama-3.1-70B-Instruct
- 71.9
Aya Expanse 32B
- 71.8 meta-llama/Meta-Llama-3-70B-Instruct
- 70.6 CIR-AMS/BTRM_Qwen2_7b_0613
- 68.8 cohere/command-r-plus-08-2024
- 68.2 allenai/tulu-2-dpo-13b
- 67.7 cohere/c4ai-aya-23-35b
- 67.3
Eurus-RM-7b
- 67.3 NousResearch/Nous-Hermes-2-Mistral-7B-DPO
- 66.8 allenai/tulu-v2.5-13b-preference-mix-rm
- 65.7
Zephyr 7B Beta
- 65.2
Aya Expanse 8B
- 63.8 meta-llama/Meta-Llama-3.1-8B-Instruct
- 63.5 c4ai-command-r-08-2024
- 62.8 meta-llama/Meta-Llama-3-8B-Instruct
- 61.0 mistralai/Mistral-7B-Instruct-v0.3
- 60.5 stabilityai/stablelm-zephyr-3b
- 59.8
Mistral-7B-Instruct-v0.2
- 58.4
Gemma 1.1 7B IT
- 1 GPT-4 Turbo Apr 2024, Generative RM 83.5%
- 2 GPT-4o May 2024, Generative RM 81.1%
- 3 google/gemma-2-9b-it Generative RM 76.5%
- 4 LxzGordon/URM-LLaMa-3.1-8B Sequence Classifier 76.3%
- 5 meta-llama/Meta-Llama-3.1-70B-Instruct Generative RM 75.5%
- 6
Aya Expanse 32B Generative RM 71.9%
- 7 meta-llama/Meta-Llama-3-70B-Instruct Generative RM 71.8%
- 8 CIR-AMS/BTRM_Qwen2_7b_0613 Sequence Classifier 70.6%
- 9 cohere/command-r-plus-08-2024 Generative RM 68.8%
- 10 allenai/tulu-2-dpo-13b DPO 68.2%
- 11 cohere/c4ai-aya-23-35b Generative RM 67.7%
- 12
Eurus-RM-7b Sequence Classifier 67.3%
- 13 NousResearch/Nous-Hermes-2-Mistral-7B-DPO DPO 67.3%
- 14 allenai/tulu-v2.5-13b-preference-mix-rm Sequence Classifier 66.8%
- 15
Zephyr 7B Beta DPO 65.7%
- 16
Aya Expanse 8B Generative RM 65.2%
- 17 meta-llama/Meta-Llama-3.1-8B-Instruct Generative RM 63.8%
- 18 c4ai-command-r-08-2024 Generative RM 63.5%
- 19 meta-llama/Meta-Llama-3-8B-Instruct Generative RM 62.8%
- 20 mistralai/Mistral-7B-Instruct-v0.3 Generative RM 61%
- 21 stabilityai/stablelm-zephyr-3b DPO 60.5%
- 22
Mistral-7B-Instruct-v0.2 Generative RM 59.8%
- 23
Gemma 1.1 7B IT Generative RM 58.4%
Source: Official M-RewardBench leaderboard
- 1 GPT-4 Turbo Apr 2024, Generative RM 83.5%
- 2 GPT-4o May 2024, Generative RM 81.1%
- 3
Aya Expanse 32B Generative RM 71.9%
- 4
Eurus-RM-7b Sequence Classifier 67.3%
- 5
Zephyr 7B Beta DPO 65.7%
- 6
Aya Expanse 8B Generative RM 65.2%
- 7 c4ai-command-r-08-2024 Generative RM 63.5%
- 8
Mistral-7B-Instruct-v0.2 Generative RM 59.8%
- 9
Gemma 1.1 7B IT Generative RM 58.4%
Source: Official M-RewardBench leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 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 | | 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 | | 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 | | 65.7% | DPO | Oct 2023 | Official leaderboard |
| 16 | | 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 | | 59.8% | Generative RM | Dec 2023 | Official leaderboard |
| 23 | | 58.4% | Generative RM | Apr 2024 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 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 | | 71.9% | Generative RM | Oct 2024 | Official leaderboard |
| 4 | | 67.3% | Sequence Classifier | Apr 2024 | Official leaderboard |
| 5 | | 65.7% | DPO | Oct 2023 | Official leaderboard |
| 6 | | 65.2% | Generative RM | Oct 2024 | Official leaderboard |
| 7 | c4ai-command-r-08-2024 | 63.5% | Generative RM | Aug 2024 | Official leaderboard |
| 8 | | 59.8% | Generative RM | Dec 2023 | Official leaderboard |
| 9 | | 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 GPT-4 Turbo Apr 2024 85.4%
- 2 GPT-4o May 2024 85%
- 3
Eurus-RM-7b ↑1 80.4%
- 4
Zephyr 7B Beta ↑1 75.5%
- 5
Aya Expanse 32B ↓2 74.5%
- 6
Aya Expanse 8B 68.2%
- 7 c4ai-command-r-08-2024 67.2%
- 8
Mistral-7B-Instruct-v0.2 65.5%
- 9
Gemma 1.1 7B IT 60.5%
Arabic
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 83.7%
- 2 GPT-4o May 2024 80.2%
- 3
Aya Expanse 32B 70.1%
- 4
Aya Expanse 8B ↑2 65%
- 5
Eurus-RM-7b ↓1 62.2%
- 6 c4ai-command-r-08-2024 ↑1 62.2%
- 7
Zephyr 7B Beta ↓2 61.3%
- 8
Mistral-7B-Instruct-v0.2 57.3%
- 9
Gemma 1.1 7B IT 56.4%
Turkish
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 82.9%
- 2 GPT-4o May 2024 79.4%
- 3
Aya Expanse 32B 70.4%
- 4
Eurus-RM-7b 63%
- 5
Aya Expanse 8B ↑1 62.8%
- 6
Zephyr 7B Beta ↓1 61.9%
- 7 c4ai-command-r-08-2024 61.5%
- 8
Mistral-7B-Instruct-v0.2 57.9%
- 9
Gemma 1.1 7B IT 56.6%
Russian
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 83.4%
- 2 GPT-4o May 2024 81%
- 3
Aya Expanse 32B 73.5%
- 4
Zephyr 7B Beta ↑1 72%
- 5
Eurus-RM-7b ↓1 69.5%
- 6
Aya Expanse 8B 65.6%
- 7 c4ai-command-r-08-2024 63.9%
- 8
Mistral-7B-Instruct-v0.2 61.9%
- 9
Gemma 1.1 7B IT 58.6%
Czech
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 83.5%
- 2 GPT-4o May 2024 80.7%
- 3
Aya Expanse 32B 73.6%
- 4
Eurus-RM-7b 68.1%
- 5
Aya Expanse 8B ↑1 66.2%
- 6
Zephyr 7B Beta ↓1 66.2%
- 7 c4ai-command-r-08-2024 63%
- 8
Mistral-7B-Instruct-v0.2 60%
- 9
Gemma 1.1 7B IT 58.7%
Polish
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 83.3%
- 2 GPT-4o May 2024 81.4%
- 3
Aya Expanse 32B 73.5%
- 4
Eurus-RM-7b 70%
- 5
Zephyr 7B Beta 67.1%
- 6
Aya Expanse 8B 64.7%
- 7 c4ai-command-r-08-2024 63.3%
- 8
Mistral-7B-Instruct-v0.2 60.8%
- 9
Gemma 1.1 7B IT 58.9%
Korean
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 82.5%
- 2 GPT-4o May 2024 79.2%
- 3
Aya Expanse 32B 67.9%
- 4
Eurus-RM-7b 65.6%
- 5
Zephyr 7B Beta 64.7%
- 6
Aya Expanse 8B 64.2%
- 7 c4ai-command-r-08-2024 60.6%
- 8
Mistral-7B-Instruct-v0.2 57.6%
- 9
Gemma 1.1 7B IT 57%
Chinese simplified
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 82.9%
- 2 GPT-4o May 2024 80.2%
- 3
Aya Expanse 32B 71.8%
- 4
Zephyr 7B Beta ↑1 68.9%
- 5
Eurus-RM-7b ↓1 68.1%
- 6
Aya Expanse 8B 65.4%
- 7 c4ai-command-r-08-2024 64.6%
- 8
Mistral-7B-Instruct-v0.2 60.8%
- 9
Gemma 1.1 7B IT 58.1%
Chinese traditional
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 83.2%
- 2 GPT-4o May 2024 79.4%
- 3
Aya Expanse 32B 73.4%
- 4
Eurus-RM-7b 68.4%
- 5
Zephyr 7B Beta 66.4%
- 6
Aya Expanse 8B 65.2%
- 7 c4ai-command-r-08-2024 63.2%
- 8
Mistral-7B-Instruct-v0.2 60.9%
- 9
Gemma 1.1 7B IT 58%
French
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 85.2%
- 2 GPT-4o May 2024 82.9%
- 3
Aya Expanse 32B 72.8%
- 4
Eurus-RM-7b 72.5%
- 5
Zephyr 7B Beta 71.5%
- 6
Aya Expanse 8B 66.2%
- 7 c4ai-command-r-08-2024 65.5%
- 8
Mistral-7B-Instruct-v0.2 61.5%
- 9
Gemma 1.1 7B IT 60%
Greek
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 82.7%
- 2 GPT-4o May 2024 81.8%
- 3
Aya Expanse 32B 69.6%
- 4
Aya Expanse 8B ↑2 64.9%
- 5 c4ai-command-r-08-2024 ↑2 61.1%
- 6
Zephyr 7B Beta ↓1 58.5%
- 7
Eurus-RM-7b ↓3 58.4%
- 8
Gemma 1.1 7B IT ↑1 57.8%
- 9
Mistral-7B-Instruct-v0.2 ↓1 55.4%
German
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 84.5%
- 2 GPT-4o May 2024 82%
- 3
Aya Expanse 32B 71.8%
- 4
Eurus-RM-7b 70.6%
- 5
Zephyr 7B Beta 70.1%
- 6
Aya Expanse 8B 67%
- 7 c4ai-command-r-08-2024 62.9%
- 8
Mistral-7B-Instruct-v0.2 61.3%
- 9
Gemma 1.1 7B IT 59.3%
Romanian
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 83.2%
- 2 GPT-4o May 2024 80.7%
- 3
Aya Expanse 32B 73.5%
- 4
Eurus-RM-7b 69.2%
- 5
Zephyr 7B Beta 65.7%
- 6 c4ai-command-r-08-2024 ↑1 64.8%
- 7
Aya Expanse 8B ↓1 64.6%
- 8
Mistral-7B-Instruct-v0.2 60.7%
- 9
Gemma 1.1 7B IT 58.7%
Italian
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 84.2%
- 2 GPT-4o May 2024 81.3%
- 3
Aya Expanse 32B 73.7%
- 4
Zephyr 7B Beta ↑1 70.9%
- 5
Eurus-RM-7b ↓1 70.4%
- 6
Aya Expanse 8B 66.5%
- 7 c4ai-command-r-08-2024 65.8%
- 8
Mistral-7B-Instruct-v0.2 62.4%
- 9
Gemma 1.1 7B IT 59.3%
Dutch
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 85.1%
- 2 GPT-4o May 2024 82.5%
- 3
Aya Expanse 32B 72.6%
- 4
Eurus-RM-7b 71.9%
- 5
Zephyr 7B Beta 69.9%
- 6
Aya Expanse 8B 66%
- 7 c4ai-command-r-08-2024 64%
- 8
Mistral-7B-Instruct-v0.2 62.8%
- 9
Gemma 1.1 7B IT 59.5%
Persian
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 81.9%
- 2 GPT-4o May 2024 80.2%
- 3
Aya Expanse 32B 68.1%
- 4
Aya Expanse 8B ↑2 65.1%
- 5 c4ai-command-r-08-2024 ↑2 60.6%
- 6
Eurus-RM-7b ↓2 59.9%
- 7
Mistral-7B-Instruct-v0.2 ↑1 56.8%
- 8
Gemma 1.1 7B IT ↑1 56.3%
- 9
Zephyr 7B Beta ↓4 55.9%
Hindi
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 83.2%
- 2 GPT-4o May 2024 79.3%
- 3
Aya Expanse 32B 70.4%
- 4
Aya Expanse 8B ↑2 62.4%
- 5
Eurus-RM-7b ↓1 62.3%
- 6 c4ai-command-r-08-2024 ↑1 61.7%
- 7
Zephyr 7B Beta ↓2 59.2%
- 8
Gemma 1.1 7B IT ↑1 58.6%
- 9
Mistral-7B-Instruct-v0.2 ↓1 55.2%
Ukrainian
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 83.1%
- 2 GPT-4o May 2024 81.4%
- 3
Aya Expanse 32B 73.9%
- 4
Eurus-RM-7b 69.6%
- 5
Zephyr 7B Beta 68.2%
- 6
Aya Expanse 8B 64.4%
- 7 c4ai-command-r-08-2024 64%
- 8
Mistral-7B-Instruct-v0.2 62.1%
- 9
Gemma 1.1 7B IT 58.7%
Portuguese
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 83.9%
- 2 GPT-4o May 2024 82.9%
- 3
Aya Expanse 32B 73%
- 4
Eurus-RM-7b 72.4%
- 5
Zephyr 7B Beta 70.9%
- 6
Aya Expanse 8B 66.3%
- 7 c4ai-command-r-08-2024 65.8%
- 8
Mistral-7B-Instruct-v0.2 62.5%
- 9
Gemma 1.1 7B IT 59.9%
Indonesian
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 83.9%
- 2 GPT-4o May 2024 82%
- 3
Aya Expanse 32B 73.6%
- 4
Eurus-RM-7b 69.1%
- 5
Zephyr 7B Beta 66.4%
- 6 c4ai-command-r-08-2024 ↑1 66.3%
- 7
Aya Expanse 8B ↓1 65.4%
- 8
Mistral-7B-Instruct-v0.2 60.3%
- 9
Gemma 1.1 7B IT 59.2%
Japanese
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 83.2%
- 2 GPT-4o May 2024 81%
- 3
Aya Expanse 32B 71.5%
- 4
Eurus-RM-7b 67.4%
- 5
Zephyr 7B Beta 65.4%
- 6
Aya Expanse 8B 65%
- 7 c4ai-command-r-08-2024 62.4%
- 8
Mistral-7B-Instruct-v0.2 58.4%
- 9
Gemma 1.1 7B IT 58.3%
Spanish
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 84.7%
- 2 GPT-4o May 2024 81.9%
- 3
Eurus-RM-7b ↑1 74%
- 4
Aya Expanse 32B ↓1 72.7%
- 5
Zephyr 7B Beta 70.9%
- 6
Aya Expanse 8B 65.8%
- 7 c4ai-command-r-08-2024 65.4%
- 8
Mistral-7B-Instruct-v0.2 64.3%
- 9
Gemma 1.1 7B IT 59%
Hebrew
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 82.4%
- 2 GPT-4o May 2024 80.6%
- 3
Aya Expanse 32B 70.5%
- 4
Aya Expanse 8B ↑2 64.2%
- 5 c4ai-command-r-08-2024 ↑2 63.1%
- 6
Eurus-RM-7b ↓2 59.7%
- 7
Zephyr 7B Beta ↓2 58.8%
- 8
Gemma 1.1 7B IT ↑1 56.9%
- 9
Mistral-7B-Instruct-v0.2 ↓1 55%
Vietnamese
9 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 84.3%
- 2 GPT-4o May 2024 82.1%
- 3
Aya Expanse 32B 72.5%
- 4
Aya Expanse 8B ↑2 66.7%
- 5
Eurus-RM-7b ↓1 66.2%
- 6 c4ai-command-r-08-2024 ↑1 65%
- 7
Zephyr 7B Beta ↓2 61.3%
- 8
Mistral-7B-Instruct-v0.2 60.5%
- 9
Gemma 1.1 7B IT 58.6%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
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.
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
- 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.