Human feedback win rates (Anthropic)
Raters chat with two models on tasks such as coding, documents and creative writing and pick the better reply. Score is the win rate.
Claude 3 era model cards report per-task win rates against a baseline Claude model, with no overall number. The baseline changes between cards.
- Made by
- Anthropic
- Released
- March 2024
- Kind
- Overall chat quality
- Level
- Everyday
- Scored
- Ranked by votes
- Answers
- Open-ended
- Tasks
- Kept private
- Last in a system card
- October 2024
- Last model added
- October 2024
5
System cards report it, from 1 lab
No scores here yet
No lab, leaderboard or Epoch AI run has reported a score for Human feedback win rates (Anthropic) yet.
Labs report it only in system cards
We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark. Some system cards do, below.
5 system cards report it, from 1 lab
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| Oct 22, 2024 | Claude 3.5 Sonnet (October 2024) model card | Human feedback win rates vs Claude 3.5 Sonnet baseline, per task with no overall: Coding 52, Creative Writing 58, Documents 61, Multilingual 48, Instruction-Following 51, Vision Instruction-Following 57 (Figure 2) | Capabilities | |
| Oct 22, 2024 | Claude 3.5 Haiku model card | Human feedback win rates vs Claude 3.5 Sonnet baseline, per task with no overall: Coding 45, Creative Writing 46, Documents 57, Multilingual 30, Instruction-Following 40. Not rated on vision (Figure 2) | Capabilities | |
| Jun 20, 2024 | Claude 3.5 Sonnet model card | Human feedback win rates vs Claude 3 Opus baseline, per task, no overall: Coding 60, Documents 66, Creative Writing 62, Multilingual 55, Instruction-Following 56, Vision Instruction-Following 61, Finance 73, Law 82, Medicine 68, Philosophy 73, STEM 58. Honesty and Harmlessness recorded separately (Figure 3) | Capabilities | |
| Mar 4, 2024 | Claude 3 Sonnet model card | Human preference win rates vs a Claude Instant 1.2 baseline, per task with no overall: Coding 69, Creative Writing 63, Instruction-following 66, Long Document Q&A 60 (Claude 2.0 baseline), Multilingual 65, Finance 80, Medicine 79, Philosophy 71, STEM 63. Honesty and Harmlessness recorded separately (Figures 5 to 7) | Capabilities | |
| Jul 11, 2023 | Claude 2 model card | Human Feedback Evaluations, task-specific Elo indexed to a baseline, no overall: Helpful 108 (Claude Instant 1.1 baseline), Harmless 278 and Honest 137 (Helpful Only 1.3 baseline) (3.1, Figure 1) | Alignment |
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).
Nemotron-Cascade 2 30B-A3B 82.9% - 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% - IFStruct In use Models generate JSON or YAML that must follow a requested schema and phrasing constraints, without constrained decoding. A response passes only if every structural constraint holds. Gemma 4 31B IT 95.9%