BenchLeader
Menu
Best LLMsChoose a modelCoding plansLeaderboardsModelsBenchmarksProvidersReleases8

FORTRESS leaderboard

As of 6 Nov 2026, the top score on FORTRESS is 74.4% by DeepSeek R1 (DeepSeek), ahead of GLM-4.5-Air at 63.2%, across 77 model configurations with a published result (Scale AI SEAL, result dated 10 Apr 2025). Paired adversarial and benign prompts across risk areas; a model must refuse the harmful ones and answer the benign ones. FORTRESS is shown for reference only and does not count toward the BenchLeader Index.

Key facts (as of 6 Nov 2026)

  • DeepSeek R1 holds the top FORTRESS score at 74.4% (Scale AI SEAL, 10 Apr 2025).
  • GLM-4.5-Air is second on FORTRESS at 63.2% (Scale AI SEAL, 13 Aug 2025).
  • FORTRESS: Robustness to harmful requests without over-refusing benign ones. Scale AI (Scale AI SEAL).
  • 77 model configurations have a published FORTRESS result on BenchLeader (BenchLeader, 6 Nov 2026).
  • FORTRESS does not feed the BenchLeader Index; it is shown for reference (BenchLeader methodology).

Robustness to harmful requests without over-refusing benign ones. Scale AI.

Published by
Scale AI SEAL
Category
Safety & honesty
Index weight
Reference only
Models
77
Data as of
6 Nov 2026
Newest result
4 Nov 2026

Scale AI SEAL Leaderboards.

What the test looks like

Paired adversarial and benign prompts across risk areas; a model must refuse the harmful ones and answer the benign ones.

How it is scored

Combined score for harmful-request refusal and benign-request compliance, published by Scale AI.

What to keep in mind

A safety measure, not a capability one; it is not part of the BenchLeader Index.

Top 15
  1. 1DeepSeek R174.4%
  2. 2GLM-4.5-Air63.2%
  3. 3Gemini 2.5 Pro61.7%
  4. 4Qwen3 235B A22B61.4%
  5. 5GLM-4.559.6%
  6. 6GPT-4.1 mini59.2%
  7. 7Qwen2.5 72B56.4%
  8. 8Mixtral 8x22B Instruct56.1%
  9. 9Kimi K255.5%
  10. 10Gemini 1.5 Pro53.9%
  11. 11GPT-4.153.0%
  12. 12MiniMax M3.1 Flash52.1%
  13. 13Gemini 3.1 Flash Lite51.0%
  14. 14Gemini 1.5 Flash50.6%
  15. 15GLM 5.248.7%
77 of 77
#
1DeepSeek R174.4%47.32025-04-10
2GLM-4.5-Air63.2%46.72025-08-13
3Gemini 2.5 Pro61.7%54.52025-04-14
4Qwen3 235B A22B61.4%50.02025-07-24
5GLM-4.559.6%51.62025-08-13
6GPT-4.1 mini59.2%44.82025-04-10
7Qwen2.5 72B56.4%43.22025-04-10
8Mixtral 8x22B Instruct56.1%37.52025-05-23
9Kimi K255.5%51.02025-07-23
10Gemini 1.5 Pro53.9%47.22025-04-10
11GPT-4.153.0%48.52025-05-23
12MiniMax M3.1 Flash52.1%53.02026-10-29
13Gemini 3.1 Flash Lite51.0%52.42026-03-13
14Gemini 1.5 Flash50.6%45.12025-04-10
15GLM 5.248.7%51.52026-11-04
16GPT-4o mini48.1%35.42025-04-10
17GPT-4o47.2%43.32025-04-10
18Llama 3.3 70B44.8%40.02025-04-10
19Llama 3.1 70B44.2%40.82025-04-10
20Gemini 3 Pro41.7%60.22025-12-02
21Kimi K2.541.1%54.22026-02-12
22Llama 4 Maverick40.1%42.12025-05-20
23Claude 3.7 Sonnet38.0%49.92025-05-24
24Claude 3.5 Haiku30.4%38.82025-05-01
25GPT 5.1 Instant30.4%–2025-11-26
26o3-mini30.1%50.72025-04-17
27Gemini 3.1 Pro29.8%60.52026-03-13
28GLM 5.3max28.2%59.02026-09-15
29Claude Opus 427.6%52.62025-04-10
30Kimi K327.3%59.12026-09-15
31DeepSeek V4.1 Flash 101727.2%–2026-10-27
32GLM 5.3 Flash26.7%56.82026-11-04
33Kimi K3max26.6%62.12026-09-15
34Nemotron 3 Ultra Nvfp4high26.4%–2026-09-15
35GPT-5.1thinking25.7%54.02025-11-26
36Inklingxhigh25.0%50.22026-09-15
37Claude Opus 4thinking24.8%55.22025-06-02
38Claude Sonnet 424.4%50.52025-04-16
39Mistral Large 423.1%52.92026-10-08
40Qwen3.8 2.4T A95Bxhigh21.6%–2026-09-15
41o4-mini21.5%56.92025-05-06
42Llama 3.1 405B20.6%41.92025-04-10
43Claude Opus 4.6no reasoning20.5%57.92026-02-17
44Muse Spark20.2%64.12026-04-08
45GPT-5.6 Sol20.1%55.02026-09-15
46o119.4%53.42025-04-16
47Claude Opus 4.8max18.2%60.22026-07-08
48GPT-6.1 Terra18.2%–2026-10-27
49Claude Sonnet 4thinking18.1%53.32025-04-16
50gpt-oss-20b17.6%44.02025-08-13
51GPT-5.217.5%57.92025-12-15
52GPT-517.0%57.72025-08-13
53GPT-5 mini17.0%56.42025-08-22
54Claude Sonnet 4.517.0%53.72025-10-02
55Claude Opus 516.8%65.22026-09-15
56GPT-5.5xhigh16.3%64.32026-07-08
57Claude Opus 4.116.1%51.42025-08-08
58o316.0%61.02025-04-16
59GPT-5 Pro15.2%59.92025-11-06
60GPT-5.4 Pro14.8%63.12026-03-23

More safety & honesty benchmarks · All LLM benchmarks

Related benchmarks

Cite as: BenchLeader, “FORTRESS leaderboard”, data as of 6 Nov 2026, https://www.benchleader.com/benchmarks/scale_fortress.

FORTRESS: questions

What does FORTRESS measure?
Paired adversarial and benign prompts across risk areas; a model must refuse the harmful ones and answer the benign ones. Scores are reported in percent of tasks solved; higher is better.
What is the top score on FORTRESS?
As of 6 Nov 2026, the highest score on FORTRESS is 74.4% by DeepSeek R1 (result dated 10 Apr 2025), ahead of GLM-4.5-Air at 63.2%. Scores come from Scale AI SEAL and are re-read daily.
Which AI model leads FORTRESS?
DeepSeek R1 leads FORTRESS with 74.4% as of 6 Nov 2026 (result dated 10 Apr 2025), ahead of GLM-4.5-Air at 63.2%.
How many models have FORTRESS results?
77 model configurations have a FORTRESS result on BenchLeader, all taken from Scale AI SEAL.
Who runs FORTRESS and how often is it updated?
FORTRESS is published by Scale AI SEAL. BenchLeader re-reads the published results once a day and records the date each result was published.
Does FORTRESS count toward the BenchLeader Index?
No. FORTRESS is shown for reference but left out of the composite index.