HarvestBench
Agents drive simulated tractors and decide how much fuel to spend avoiding animals. The benchmark measures how often they hit animals as avoidance costs change.
- Made by
- Jasmine Brazilek and colleagues
- Released
- September 2026
- Kind
- Deception and alignment
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- All public
- Measures
- Propensity
- Behavior
- Misaligned with values
- Language
- English
- Interaction
- Agentic
- Environment
- A simulator
- Items from
- Generated
- Refresh
- One fixed set
- Run by
- Anyone
- Strongest model tested
- GPT-6 Astra #4 on the capability index
- Last model added
- September 2026
- Scores refreshed
- November 20th, 2026
GPT-5.6 Terra leads with 0.40% among models released by September 2026 GPT-5.6 Terra leads with 0.40% among models released by September 2026
Best score per entry from the benchmark's own leaderboard, read on November 20th, 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 September 2026, 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
- Claude Opus 5.5
- and 13 more
- 1 GPT-5.6 Terra Morality briefing 0.40%
- 2 GPT-6 Astra Morality briefing 0.50%
- 3 GPT-5.6 Sol Morality briefing 0.90%
- 4
DeepSeek-V3.1 Morality briefing 2.40%
- 5 Claude Haiku 4.5 Morality briefing 4.50%
- 6 GPT-5 mini Morality briefing 5.40%
- 7 Claude Sonnet 5 Morality briefing 17.8%
- 8
Gemini 2.5 Flash Sep 2025, Morality briefing 38.7%
- 9
Mistral Small 3.2 Morality briefing 88.8%
- 10 GPT-4o-mini (snapshot unspecified) Morality briefing 98.8%
Source: Official HarvestBench leaderboard
- 1 GPT-5.6 Terra Morality briefing 0.40%
- 2 GPT-6 Astra Morality briefing 0.50%
- 3 GPT-5.6 Sol Morality briefing 0.90%
- 4
DeepSeek-V3.1 Morality briefing 2.40%
- 5 Claude Haiku 4.5 Morality briefing 4.50%
- 6 GPT-5 mini Morality briefing 5.40%
- 7 Claude Sonnet 5 Morality briefing 17.8%
- 8
Gemini 2.5 Flash Sep 2025, Morality briefing 38.7%
- 9
Mistral Small 3.2 Morality briefing 88.8%
Source: Official HarvestBench leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5.6 Terra | 0.40% | Morality briefing | Jul 2026 | Official leaderboard |
| 2 | GPT-6 Astra | 0.50% | Morality briefing | Sep 2026 | Official leaderboard |
| 3 | GPT-5.6 Sol | 0.90% | Morality briefing | Jul 2026 | Official leaderboard |
| 4 | | 2.40% | Morality briefing | Aug 2025 | Official leaderboard |
| 5 | Claude Haiku 4.5 | 4.50% | Morality briefing | Oct 2025 | Official leaderboard |
| 6 | GPT-5 mini | 5.40% | Morality briefing | Aug 2025 | Official leaderboard |
| 7 | Claude Sonnet 5 | 17.8% | Morality briefing | Jun 2026 | Official leaderboard |
| 8 | | 38.7% | Morality briefing | Sep 2025 | Official leaderboard |
| 9 | | 88.8% | Morality briefing | Jun 2025 | Official leaderboard |
| 10 | GPT-4o-mini (snapshot unspecified) | 98.8% | Morality briefing | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5.6 Terra | 0.40% | Morality briefing | Jul 2026 | Official leaderboard |
| 2 | GPT-6 Astra | 0.50% | Morality briefing | Sep 2026 | Official leaderboard |
| 3 | GPT-5.6 Sol | 0.90% | Morality briefing | Jul 2026 | Official leaderboard |
| 4 | | 2.40% | Morality briefing | Aug 2025 | Official leaderboard |
| 5 | Claude Haiku 4.5 | 4.50% | Morality briefing | Oct 2025 | Official leaderboard |
| 6 | GPT-5 mini | 5.40% | Morality briefing | Aug 2025 | Official leaderboard |
| 7 | Claude Sonnet 5 | 17.8% | Morality briefing | Jun 2026 | Official leaderboard |
| 8 | | 38.7% | Morality briefing | Sep 2025 | Official leaderboard |
| 9 | | 88.8% | Morality briefing | Jun 2025 | Official leaderboard |
Rankings by part
We read 3 parts from the benchmark’s own leaderboard: Neutral briefing animal continue rate, Hay continue rate and Rock continue rate. Each ranks models on one part of the test. GPT-5.6 Sol leads 1 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Neutral briefing animal continue rate
7 models we track have a score here.
- 1 GPT-5.6 Sol ↑2 84.6%
- 2 GPT-5 mini ↑4 91%
- 3
Mistral Small 3.2 ↑6 92.2%
- 4 Claude Haiku 4.5 ↑1 98.4%
- 5
DeepSeek-V3.1 ↓1 100%
- 6
Gemini 2.5 Flash Sep 2025 ↑2 100%
- 7 GPT-5.6 Terra ↓6 100%
Hay continue rate
9 models we track have a score here.
- 1 GPT-6 Astra ↑1 0.40%
- 2 GPT-5.6 Sol ↑1 1.40%
- 3 GPT-5.6 Terra ↓2 70.3%
- 4
DeepSeek-V3.1 91.8%
- 5 Claude Haiku 4.5 95.1%
- 6 GPT-5 mini 96.6%
- 7
Mistral Small 3.2 ↑2 100%
- 8
Gemini 2.5 Flash Sep 2025 100%
- 9 Claude Sonnet 5 ↓2 100%
Rock continue rate
9 models we track have a score here.
- 1
Mistral Small 3.2 ↑8 0%
- 2 GPT-5 mini ↑4 0%
- 3
DeepSeek-V3.1 ↑1 0%
- 4 Claude Haiku 4.5 ↑1 0%
- 5 Claude Sonnet 5 ↑2 0%
- 6 GPT-5.6 Terra ↓5 0%
- 7 GPT-6 Astra ↓5 0%
- 8 GPT-5.6 Sol ↓5 0.30%
- 9
Gemini 2.5 Flash Sep 2025 ↓1 0.90%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official HarvestBench 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 HarvestBench 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 Safety and alignment
- FORTRESS In use Paired adversarial and benign national-security misuse prompts. Rubric-graded risk score on adversarial and over-refusal score on benign prompts. Claude Opus 5.5 5.78%
- StrongREJECT In use Forbidden-prompt jailbreak benchmark with a rubric grader that scores how useful a non-refusal actually is.
Inkling 98.6%
- Gray Swan IPI Arena benchmark In use Indirect prompt-injection attacks from public red-teaming competitions replayed on tool, coding and browser agents. Attack success rate at k attempts.
Gemini 4 Carbon 0%
- Covert misbehavior (Thinking Machines) In use Agentic tasks that check whether a model misbehaves and conceals it. Scored by the rate of covert misbehavior.
Inkling 2 Mini 8.30%
- Destructive actions (Thinking Machines) In use Agentic tasks that check whether a model takes destructive actions it was not authorized to take. Scored by the rate of destructive actions.
Inkling 2 Small 4.70%
- Misbehavior (Thinking Machines) In use Agentic tasks that check whether a model fabricates results, ignores instructions or hides what it did when a task is hard. Scored by the rate of misbehavior.
Inkling 2 Mini 9.80%
Data audit
All 9 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (10 entries).
- Pass Models scored 9 models we track, plus 1 other entry on its leaderboard.
- Pass Newest score The newest is from September 2026, for GPT-6 Astra.
- Note Frontier models Only 1 of the top 10 has a score. The strongest tested is GPT-6 Astra (#4).
- Note Headroom Lower is better, so there’s no ceiling to reach.
Leaderboard import 3 pass
- Pass Importer We read the board on compassionbench.com: 9 rows for models we track and 1 other entry 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.