Skip to content

The daily check, model by model.

36 of 39 listed models answered today's fixed prompt the way that build answers it. Generated 2026-11-11 16:25 UTC.

daily check, 2026-11-11

36of 39 steady

One cell per listed model, in name order; hover a cell for its verdict.

verdicts today
Listed models
39
Steady
36
Changed
0
Not checked yet
3
steady the longest
curl
curl https://pentestllm.ai/proof.json
curl https://pentestllm.ai/proof.csv
Daily real-model check
Abliterated Large FlashSteady since Oct 30Nov 11Oct 30
Abliterated ModelSteady since Oct 30Nov 11Oct 30
Abliterated Model LargeSteady since Oct 30Nov 11Oct 30
Abliterated Model Large V2Steady since Nov 4Nov 9Nov 4
Apex Flash 1 AbliteratedSteady since Nov 4Nov 11Nov 4
DeepSeek R1 Distill Llama 70B AbliteratedSteady since Oct 30Oct 30Oct 30
DeepSeek R1 Distill Qwen 32B AbliteratedSteady since Oct 23Nov 3Oct 23
DeepSeek V4 Pro AbliteratedSteady since Nov 6Nov 11Nov 6
DeepSeek V4 Pro Qwen3.5 9B Uncensored HereticSteady since Nov 8Nov 11Nov 8
DeepSeek V4.1 Flash 1017 AbliteratedSteady since Nov 6Nov 11Nov 6
DeepSeek V4.1 Flash 1017 HereticSteady since Nov 6Nov 11Nov 6
DeepSeek V4.1 Flash 1017 Heretic ThinkingSteady since Nov 6Nov 11Nov 6
DeepSeek V4.1 Flash AbliteratedSteady since Oct 19Nov 11Oct 19
DeepSeek V4.1 Flash Abliterated FastSteady since Oct 25Nov 11Oct 25
Gemma 31B Heretic UncensoredSteady since Oct 23Nov 11Oct 23
Gemma 4 26B A4B CybersecuritySteady since Oct 29Nov 11Oct 29
Gemma 4 31B Cognitive UnshackledSteady since Oct 23Nov 6Oct 23
GLM-4.6 Derestricted v5Not checked yetn/an/a
GLM-5.3 AbliteratedSteady since Oct 29Nov 11Oct 29
GLM-5.3 Flash AbliteratedSteady since Oct 29Nov 11Oct 29
GLM-5.3 Flash Apex AbliteratedSteady since Oct 23Nov 11Oct 23
GLM-5.3 Flash CybersecuritySteady since Oct 23Nov 11Oct 23
Llama 3 70B Abliterated v3.5Steady since Oct 29Nov 11Oct 29
Llama 3.3 70B AbliteratedSteady since Oct 29Nov 11Oct 29
MiMo V2.6 Flash AbliteratedSteady since Oct 24Nov 11Oct 24
NeuralDaredevil 8B AbliteratedSteady since Oct 24Nov 11Oct 24
OrcaCyber MiniMax M3.1 FlashNot checked yetn/an/a
OrcaCyber MiniMax M3.1 FlashNot checked yetn/an/a
OrcaCyber Qwen3.8 Flash NextSteady since Nov 8Nov 11Nov 8
OrcaCyber Qwen3.8 Flash NextSteady since Nov 6Nov 11Nov 6
OrcaCyber Zero 1.5Steady since Nov 7Nov 11Nov 7
Qwen 2.5 32B AbliteratedSteady since Oct 30Nov 11Oct 30
Qwen3.5 27B Blossom V6.4 DerestrictedSteady since Oct 31Nov 11Oct 31
Qwen3.5 27B BlueStar v3 DerestrictedSteady since Oct 23Nov 10Oct 23
Qwen3.5 27B Claude 4.6 Opus Reasoning Distilled DerestrictedSteady since Oct 23Nov 9Oct 23
Qwen3.5 27B Claude 4.6 Opus Reasoning Distilled Derestricted LiteSteady since Oct 30Nov 9Oct 30
Qwen3.5 27B Queen DerestrictedSteady since Oct 23Nov 10Oct 23
Qwen3.8 27B CybersecuritySteady since Oct 29Nov 11Oct 29
Qwen3.8 27B ObliteratedSteady since Oct 22Nov 11Oct 22

What the check tests

Once a day every model on sale gets the same fixed test prompt and is asked for a single output token. The input token count comes from the model’s own tokenizer, so a genuine build reports the same count day after day. A swapped model, or instructions slipped into the prompt on the way, shows up as a different count.

It is evidence, not proof: two builds that share a tokenizer report the same count, so the check catches a different model or an added prompt, not a swap between two tunes of one base. The same rows are public as JSON and CSV.