probity. deterministic model integrity.

every gguf model audited here was decomposed at the byte level. content-addressed fingerprints, per-attention-head and per-moe-expert decomposition, signed regulatory claims, replay-verified by the public probity binary.

9 models indexed · 68 signed claims · 1.15T parameters in catalog
ⓘ first time here? what am i looking at?

this page is the public catalog of probity audit dossiers — one card per audited gguf model. probity is optrenium.ai's proprietary model-audit platform that takes any gguf file and produces a deterministic, byte-level audit dossier: cryptographic fingerprints, per-attention-head and per-moe-expert decomposition, statistical outlier scans, tokenizer security checks, and a single 0–100 deploy-readiness score. the audit reports below are published free for public reference; the probity tool itself is available under commercial license.

each card shows the model's headline numbers + the readiness score with its tier (clean / advisory / caution / blocked). click a card to open the full dossier, which is a 13-tab spa with everything probity found, plus an ⓘ what is this? info panel at the top of every tab.

every claim is replay-verifiable: anyone with the original gguf bytes + the same probity binary version can re-run the audit and confirm the claim ids match byte-for-byte. no trust required — that's the whole point.

tier
family
shape
0 shown

unsloth/GLM-5.1-GGUF

advisory · 75 moe · 256/layer split · 17 shards huggingface.co
glm-dsa ·256x22B ·mit · arch fp 68758f7163fce2e9…
753.86B
parameters
42.38B active/token
tensors
1809
blocks
79
quant
Q8_0 +1
anomalies
5 hampel outliers 582 outlier-channel tensors (worst 631.9×)
8 signed claims · 746.31 GiB across 17 shards 2026-05-15 13:07 utc

google/gemma-4-31b

advisory · 61 lmstudio.ai
gemma4 ·31B · arch fp 14a8954e4841779d…
30.70B
parameters
tensors
833
blocks
60
quant
Q8_0 +1
anomalies
41m
47 hampel outliers 48 outlier-channel tensors (worst 15.5×)
7 signed claims · 30.39 GiB 2026-05-14 05:52 utc

google/gemma-4-26b-a4b

advisory · 71 moe · 128/layer lmstudio.ai
gemma4 ·26B-A4B · arch fp 7c9d9c30da408afb…
25.23B
parameters
3.82B active/token
tensors
658
blocks
30
quant
Q8_0 +1
anomalies
41m
22 hampel outliers 18 outlier-channel tensors (worst 8.4×)
8 signed claims · 25.02 GiB 2026-05-14 05:52 utc

google/gemma-4-e2b

caution · 58 lmstudio.ai
gemma4 ·4.6B · arch fp 7430f29964a17272…
4.65B
parameters
tensors
601
blocks
35
quant
Q8_0 +2
anomalies
41m
33 hampel outliers 44 outlier-channel tensors (worst 17.3×)
7 signed claims · 4.63 GiB 2026-05-14 05:52 utc

google/gemma-4-e4b

caution · 59 lmstudio.ai
gemma4 ·7.5B · arch fp b8d248467b76a9ff…
7.52B
parameters
tensors
720
blocks
42
quant
Q8_0 +2
anomalies
41m
31 hampel outliers 36 outlier-channel tensors (worst 69.9×)
7 signed claims · 7.48 GiB 2026-05-14 05:51 utc

minimax/minimax-m2.7

clean · 86 moe · 256/layer split · 7 shards lmstudio.ai
minimax-m2 ·256x4.9B · arch fp 8d3e148b5c6f1ce1…
228.69B
parameters
11.03B active/token
tensors
809
blocks
62
quant
Q8_0 +1
anomalies
6 hampel outliers 61 outlier-channel tensors (worst 21.5×)
8 signed claims · 226.44 GiB across 7 shards 2026-05-14 05:51 utc

qwen/qwen3.6-35b-a3b

clean · 87 moe · 256/layer lmstudio.ai
qwen35moe ·35B-A3B · arch fp a9d345561e1112b2…
34.66B
parameters
3.45B active/token
tensors
733
blocks
40
quant
Q8_0 +1
anomalies
1m
1 collapsed experts 154 outlier-channel tensors (worst 28.7×)
7 signed claims · 34.37 GiB 2026-05-14 05:49 utc

nvidia/nemotron-3-nano

clean · 85 moe · 128/layer lmstudio.ai
nemotron_h_moe ·30B · arch fp 595007ad917098df…
31.58B
parameters
3.58B active/token
tensors
401
blocks
52
quant
Q8_0 +1
anomalies
2m
12 hampel outliers 94 outlier-channel tensors (worst 23.5×)
8 signed claims · 31.28 GiB 2026-05-14 05:49 utc

nvidia/nemotron-3-nano-omni

clean · 85 moe · 128/layer lmstudio.ai
nemotron_h_moe ·30B-A3B · arch fp 07d6a7748690fff6…
31.58B
parameters
3.58B active/token
tensors
401
blocks
52
quant
Q8_0 +1
anomalies
2m
11 hampel outliers 94 outlier-channel tensors (worst 23.5×)
8 signed claims · 31.28 GiB 2026-05-14 05:48 utc

every audit on this page was generated by probity, optrenium.ai's proprietary model-audit platform. the audit reports themselves are replay-verifiable — when run against the same gguf bytes by a licensed probity instance, every claim id re-derives byte-identical.

open-source gguf models are audited and the reports hosted on this site free of charge*.

* probity itself is a commercial product. private & closed-source providers can license probity for in-house use — licensing@optrenium.ai

glossary · hover any underlined term for its definition