# HoneyBench leaderboard

external benchmark

> Every figure below is reproduced as its upstream source published it: nothing is modelled, estimated, interpolated or converted. `Unavailable` means no source published the value — it is never a zero. Each value carries its evidence state and the date it was observed.

Page: https://modelscale.dev/leaderboards/benchmarks/external-honey-bench

HoneyBench v0.2 (Goodhart Labs). Every model the catalog carries a published HoneyBench value for, ranked by that value.

## Benchmark record

| Field | Value |
| --- | --- |
| Category | external |
| Measure | Reward-hacking rate (%, lower is better) |
| Tasks | 20 private honeypot tasks across 7 domains with exploitable evaluators |
| Difficulty | Agentic alignment and specification gaming |
| Published by | [HoneyBench v0.2](https://goodhartlabs.com/honeybench) |

A reward-hacking and specification-gaming safety benchmark from Goodhart Labs that pairs difficult agent tasks with exploitable graders to measure how often models cheat instead of solving the task honestly. Lower scores are better.

## HoneyBench ranking

21 models with a published HoneyBench value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.

| Rank | Model | Provider | Reward-hacking rate (%, lower is better) | Evidence |
| --- | --- | --- | --- | --- |
| 1 | [Muse Spark Open](https://modelscale.dev/models/muse-spark-open) | Meta | 85.6 | Observed 2026-11-21 · source benchlm:models |
| 2 | [Grok 4.7](https://modelscale.dev/models/grok-4-7) | xAI | 85.1 | Observed 2026-11-21 · source benchlm:models |
| 3 | [Mistral Large 4](https://modelscale.dev/models/mistral-large-4) | Mistral | 83.6 | Observed 2026-11-21 · source benchlm:models |
| 3 | [Muse Spark 1.3](https://modelscale.dev/models/muse-spark-1-3) | Meta | 83.6 | Observed 2026-11-21 · source benchlm:models |
| 5 | [Muse Wonder](https://modelscale.dev/models/muse-wonder) | Meta | 81.0 | Observed 2026-11-21 · source benchlm:models |
| 6 | [Gemini 3.8 Flash](https://modelscale.dev/models/gemini-3-8-flash) | Google | 80.5 | Observed 2026-11-21 · source benchlm:models |
| 7 | [Kimi K3](https://modelscale.dev/models/kimi-3) | Moonshot AI | 78.5 | Observed 2026-11-21 · source benchlm:models |
| 8 | [GLM-5.3](https://modelscale.dev/models/glm-5-3) | Z.AI | 76.8 | Observed 2026-11-21 · source benchlm:models |
| 9 | [DeepSeek V4.1 Flash 1017](https://modelscale.dev/models/deepseek-v4-1-flash) | DeepSeek | 76.4 | Observed 2026-11-21 · source benchlm:models |
| 10 | [Kimi K3.1 Pro](https://modelscale.dev/models/kimi-3-1-pro) | Moonshot AI | 71.8 | Observed 2026-11-21 · source benchlm:models |
| 11 | [DeepSeek V4 Pro](https://modelscale.dev/models/deepseek-v4-pro) | DeepSeek | 71.4 | Observed 2026-11-21 · source benchlm:models |
| 12 | [Qwen3.8 Max](https://modelscale.dev/models/qwen3-8-max) | Alibaba | 66.2 | Observed 2026-11-21 · source benchlm:models |
| 13 | [Step 5 Preview](https://modelscale.dev/models/step-5-preview) | StepFun | 64.9 | Observed 2026-11-21 · source benchlm:models |
| 14 | [Gemini 4 Argon](https://modelscale.dev/models/gemini-4-argon) | Google | 63.6 | Observed 2026-11-21 · source benchlm:models |
| 15 | [Claude Fable 5.1](https://modelscale.dev/models/claude-fable-5-1) | Anthropic | 61.8 | Observed 2026-11-21 · source benchlm:models |
| 16 | [GPT-6.1 Sol](https://modelscale.dev/models/gpt-6-1-sol) | OpenAI | 59.5 | Observed 2026-11-21 · source benchlm:models |
| 17 | [Inkling 2 Small](https://modelscale.dev/models/inkling-2-small) | Thinking Machines Lab | 52.6 | Observed 2026-11-21 · source benchlm:models |
| 18 | [GPT-6 Astra](https://modelscale.dev/models/gpt-6-astra) | OpenAI | 50.8 | Observed 2026-11-21 · source benchlm:models |
| 19 | [Claude Sonnet 5.5](https://modelscale.dev/models/claude-sonnet-5-5) | Anthropic | 41.6 | Observed 2026-11-21 · source benchlm:models |
| 20 | [Claude Opus 5.5](https://modelscale.dev/models/claude-opus-5-5) | Anthropic | 31.6 | Observed 2026-11-21 · source benchlm:models |
| 21 | [Claude Fable 5.2](https://modelscale.dev/models/claude-fable-5-2) | Anthropic | 11.9 | Observed 2026-11-21 · source benchlm:models |

Rows are ordered by the value the source published, highest first. The source does not state whether a higher value is the better result, so this document does not either: for a benchmark that measures a rate of failure, read the table from the bottom.
