Benchmark Almanac

Skill Reward Hacking

Qwen3.6 27B 13.8 leads since May 2026
Not in model cards

Researchers track when models learn to increase a proxy reward without improving task quality. They measure warning time, false alarms, and transfer across reward functions.

The highest public hub mean reward was 13.76 for qwen3.6-27b on 50 examples on June 16th, 2026. The hub also lists mimo-v2.5-pro, Llama-3.2-1B-Instruct, glm-5.1.

Made by
tonyteo
Released
May 2026
Kind
Deception and alignment
Level
University
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Measures
Propensity
Language
English
In and out
Text → Actions
Interaction
Agentic
Refresh
One fixed set
Run by
Anyone
Strongest model tested
GLM-5.1 #69 on the capability index
Last model added
April 2026
Scores refreshed
November 20th, 2026
5
Entries on its own leaderboard, 4 models we track

Qwen3.6 27B leads with 13.8 among models released by April 2026 Qwen3.6 27B leads with 13.8 among models released by April 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 April 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 Qwen3.6 27B 9.5.0, 50 examples 13.8
  2. 2 MiMo-V2.5-Pro 2.0.0, 10 examples 9.48
  3. 3 meta-llama/Llama-3.2-1B-Instruct 2.0.0, 8 examples 8.96
  4. 4 GLM-5.1 1.0.0, 8 examples 5.24
  5. 5 GPT-5 mini 1.0.0, 8 examples 4.95

Source: Official Skill Reward Hacking leaderboard

#EntryScoreSettingSubmitted byReleasedRun onLinkSource
1 Qwen3.6 27B 13.8 9.5.0, 50 examples Nguyen An Apr 2026 Jun 2026 app.primeintellect.ai Official leaderboard
2 MiMo-V2.5-Pro 9.48 2.0.0, 10 examples Hung Trinh Apr 2026 May 2026 app.primeintellect.ai Official leaderboard
3 meta-llama/Llama-3.2-1B-Instruct 8.96 2.0.0, 8 examples Hung Trinh May 2026 app.primeintellect.ai Official leaderboard
4 GLM-5.1 5.24 1.0.0, 8 examples Hung Trinh Apr 2026 May 2026 app.primeintellect.ai Official leaderboard
5 GPT-5 mini 4.95 1.0.0, 8 examples Hung Trinh Aug 2025 May 2026 app.primeintellect.ai Official leaderboard
#ModelScoreSettingSubmitted byReleasedRun onLinkSource
1 Qwen3.6 27B 13.8 9.5.0, 50 examples Nguyen An Apr 2026 Jun 2026 app.primeintellect.ai Official leaderboard
2 MiMo-V2.5-Pro 9.48 2.0.0, 10 examples Hung Trinh Apr 2026 May 2026 app.primeintellect.ai Official leaderboard
3 GLM-5.1 5.24 1.0.0, 8 examples Hung Trinh Apr 2026 May 2026 app.primeintellect.ai Official leaderboard
4 GPT-5 mini 4.95 1.0.0, 8 examples Hung Trinh Aug 2025 May 2026 app.primeintellect.ai Official leaderboard

Number one over time

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

Qwen
Own leaderboard Not tested since May 2026

Source: Official Skill Reward Hacking 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
4.00 6.00 8.00 10.0 12.0 14.0 2026 4.00 6.00 8.00 10.0 12.0 14.0 2026

Source: Official Skill Reward Hacking 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

All 636 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

8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 7 months ago, in April 2026, for MiMo-V2.5-Pro. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
  • Pass Independent runs Scores from the official leaderboard (5 entries).
  • Pass Models scored 4 models we track, plus 1 other entry on its leaderboard.
  • Warning Newest score The newest came 7 months ago, in April 2026, for MiMo-V2.5-Pro.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GLM-5.1 (#69).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on app.primeintellect.ai: 5 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.

How we check