Benchmark Almanac

MCPTox

Qwen3-14B 5.10% leads since August 2025
Not in model cards

Agents use tools whose descriptions contain untrusted instructions. The benchmark measures whether agents follow those instructions across 1,348 cases from 45 servers.

The August 2025 paper reports 5.1% attack success for Qwen3-14B without reasoning, compared with 72.8% for o1-mini. Lower is better.

Made by
MCPTox authors
Released
August 2025
Kind
Jailbreaks and prompt injection
Level
University
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Measures
Propensity
Language
English
In and out
Text → Actions
Interaction
Agentic
Environment
APIs and tools
Refresh
One fixed set
Run by
Anyone
Strongest model tested
Claude 3.7 Sonnet #119 on the capability index
Last model added
April 2025
Scores refreshed
November 20th, 2026
16
Entries in its paper, 9 models we track

Qwen3-14B leads with 5.10% among models released by April 2025 Qwen3-14B leads with 5.10% among models released by April 2025

Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. 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 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 Qwen3-14B Reasoning disabled 5.10%
  2. 2 Mistral Reasoning disabled 8.30%
  3. 3 Qwen3-8B Reasoning disabled 14%
  4. 4 Llama 3.1-8B Reasoning disabled 14.1%
  5. 5 Gemma 2 9B Reasoning disabled 14.5%
  6. 6 GPT-3.5-turbo (snapshot unspecified) Reasoning disabled 14.9%
  7. 7 Qwen3-235B-A22B Reasoning disabled 17.2%
  8. 8 Qwen3-32B Reasoning disabled 23.7%
  9. 9 Llama 3.1-70B Reasoning disabled 24.6%
  10. 10 Claude 3.7 Sonnet Reasoning enabled 34.3%
  11. 11 DeepSeek-V3 (snapshot unspecified) Reasoning disabled 56.5%
  12. 12 Gemini-2.5-flash (snapshot unspecified) Reasoning enabled 59.7%
  13. 13 GPT-4o-mini (snapshot unspecified) Reasoning disabled 61.8%
  14. 14 Phi-4-Think Reasoning enabled 70.2%
  15. 15 DeepSeek-R1 (snapshot unspecified) Reasoning enabled 70.9%
  16. 16 o1-mini Reasoning enabled 72.8%

Source: The MCPTox paper

#EntryScoreSettingDateSource
1 Qwen3-14B 5.10% Reasoning disabled Apr 2025 Official leaderboard
2 Mistral 8.30% Reasoning disabled Official leaderboard
3 Qwen3-8B 14% Reasoning disabled Apr 2025 Official leaderboard
4 Llama 3.1-8B 14.1% Reasoning disabled Jul 2024 Official leaderboard
5 Gemma 2 9B 14.5% Reasoning disabled Jun 2024 Official leaderboard
6 GPT-3.5-turbo (snapshot unspecified) 14.9% Reasoning disabled Official leaderboard
7 Qwen3-235B-A22B 17.2% Reasoning disabled Apr 2025 Official leaderboard
8 Qwen3-32B 23.7% Reasoning disabled Apr 2025 Official leaderboard
9 Llama 3.1-70B 24.6% Reasoning disabled Jul 2024 Official leaderboard
10 Claude 3.7 Sonnet 34.3% Reasoning enabled Feb 2025 Official leaderboard
11 DeepSeek-V3 (snapshot unspecified) 56.5% Reasoning disabled Official leaderboard
12 Gemini-2.5-flash (snapshot unspecified) 59.7% Reasoning enabled Official leaderboard
13 GPT-4o-mini (snapshot unspecified) 61.8% Reasoning disabled Official leaderboard
14 Phi-4-Think 70.2% Reasoning enabled Official leaderboard
15 DeepSeek-R1 (snapshot unspecified) 70.9% Reasoning enabled Official leaderboard
16 o1-mini 72.8% Reasoning enabled Sep 2024 Official leaderboard
#ModelScoreSettingDateSource
1 Qwen3-14B 5.10% Reasoning disabled Apr 2025 Official leaderboard
2 Qwen3-8B 14% Reasoning disabled Apr 2025 Official leaderboard
3 Llama 3.1-8B 14.1% Reasoning disabled Jul 2024 Official leaderboard
4 Gemma 2 9B 14.5% Reasoning disabled Jun 2024 Official leaderboard
5 Qwen3-235B-A22B 17.2% Reasoning disabled Apr 2025 Official leaderboard
6 Qwen3-32B 23.7% Reasoning disabled Apr 2025 Official leaderboard
7 Llama 3.1-70B 24.6% Reasoning disabled Jul 2024 Official leaderboard
8 Claude 3.7 Sonnet 34.3% Reasoning enabled Feb 2025 Official leaderboard
9 o1-mini 72.8% Reasoning enabled Sep 2024 Official leaderboard

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Qwen3-32B
Reasoning d… Reasoning e…
58.5%
Qwen3-235B-A22B
Reasoning d… Reasoning e…
50.6%
Qwen3-8B
Reasoning d… Reasoning e…
41.8%
Qwen3-14B
Reasoning d… Reasoning e…
27.1%

Source: The MCPTox paper

ModelEffortHarnessScoreCost per taskOutput tokens
o1-mini Reasoning enabled 72.8%
Qwen3-32B Reasoning enabled 58.5%
Qwen3-235B-A22B Reasoning enabled 50.6%
Qwen3-8B Reasoning enabled 41.8%
Claude 3.7 Sonnet Reasoning enabled 34.3%
Qwen3-14B Reasoning enabled 27.1%
Llama 3.1-70B Reasoning disabled 24.6%
Qwen3-32B Reasoning disabled 23.7%
Qwen3-235B-A22B Reasoning disabled 17.2%
Gemma 2 9B Reasoning disabled 14.5%
Llama 3.1-8B Reasoning disabled 14.1%
Qwen3-8B Reasoning disabled 14%
Qwen3-14B Reasoning disabled 5.10%

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Qwen3-32B
Reasoning d… Reasoning e…
58.5%
Qwen3-235B-A22B
Reasoning d… Reasoning e…
50.6%
Qwen3-8B
Reasoning d… Reasoning e…
41.8%
Qwen3-14B
Reasoning d… Reasoning e…
27.1%

Source: The MCPTox paper

ModelEffortHarnessScoreCost per taskOutput tokens
o1-mini Reasoning enabled 72.8%
Qwen3-32B Reasoning enabled 58.5%
Qwen3-235B-A22B Reasoning enabled 50.6%
Qwen3-8B Reasoning enabled 41.8%
Claude 3.7 Sonnet Reasoning enabled 34.3%
Qwen3-14B Reasoning enabled 27.1%
Llama 3.1-70B Reasoning disabled 24.6%
Qwen3-32B Reasoning disabled 23.7%
Qwen3-235B-A22B Reasoning disabled 17.2%
Gemma 2 9B Reasoning disabled 14.5%
Llama 3.1-8B Reasoning disabled 14.1%
Qwen3-8B Reasoning disabled 14%
Qwen3-14B Reasoning disabled 5.10%

Number one over time

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

Qwen
Benchmark paper Not tested since Aug 2025
Qwen3-14B 5.10%

Source: The MCPTox paper

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
In the benchmark’s paper Best so far
0% 20% 40% 60% 80% 100% 2024 2025 Qwen3-14B Qwen3-8B Llama 3.1-8B 0% 20% 40% 60% 80% 100% 2024 2025 Qwen3-14B Qwen3-8B

Source: Official MCPTox 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 19 months ago, in April 2025, for Qwen3-14B. 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 benchmark’s paper (16 entries).
  • Pass Models scored 9 models we track, plus 7 other entries on its leaderboard.
  • Warning Newest score The newest came 19 months ago, in April 2025, for Qwen3-14B.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Claude 3.7 Sonnet (#119).
  • Note Headroom Lower is better, so there’s no ceiling to reach.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 13 rows for models we track and 7 other entries 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