OpenVals

The Trust Layer for AI

OpenVals helps enterprises validate, secure, and trust AI before production.

pip install openvals
OpenVals DocsView on GitHub
150+
validated models
5+
LLM Providers
20+
Core Metrics
v0.5.5
Current Release
Dr.Pinnacle
License
Independent Evaluation, Unbiased Benchmarks

Testing AI on Real-World Tasks

We benchmark the world's leading AI models on economically valuable tasks such as finance, software, and frontier risk like cybersecurity, recursive self improvement and mental health. We run all of our own evaluations and create many of our benchmarks in-house.

OPENVALS INDEX•OCT 13, 2026
100.0%
DEEPSCALER1.5B
98.3%
HUNYUAN1.8B
96.7%
GRANITE 3.32.0B
95.0%
QWEN 30.6B
91.4%
TERNARYBONSAI 1.7B
90.8%
LFM 2.5THINKING
90.2%
MEDPSY1.7B
83.9%
SMOLLM1.7B
82.5%
GEMMA 3NE2B
80.5%
LLAMA 2.01.0B
78.9%
PHI-33.8B
77.5%
FALCON 31.0B
76.2%
OPENCHAT7B
69.8%
YI-CODER1.5B
69.6%
DEEPSEEKR1-QWEN 1.5B
67.4%
MINISTRAL3.0B
66.7%
STABLELMZEPHYR 3B
56.7%
HERMES 33.0B
47.2%
ONELLMDOEY 1.0B
37.7%
BLOOM560M
37.3%
SUPRA50M
19.7%
ASENAESP32 121M
12.4%
WILLOWALPHA 0.3B
iShowing evaluation models in descending order
View Full Results

Latest Reports

Recent benchmark releases and model evaluations across open-weight and enterprise AI.

Validated Model Registry (70)

Latest Openvals Validation Metrics
MODELSPerformanceTrustRANKINGS
AccuracySemanticLatencyFactualityHallucinationSafetyReliabilityDRS
granite3-dense2.0b1.6 GB
76.8%
67.6%3243 ms74.1%52.5%100.0%94.3%64.4%1/70
gemma2.0b1.7 GB
79.2%
66.6%5511 ms73.3%53.4%100.0%87.9%63.2%2/70
openchat:latest7b4.1GB
76.2%
67.3%43021 ms73.8%53.6%100.0%92.8%63.1%3/70
lfm2.5-thinking1.2b731 MB
90.8%
52.6%9364 ms61.8%59.0%94.8%93.7%61.6%4/70
granite3.32.0b1.5 GB
96.7%
44.2%2428 ms55.4%61.2%100.0%92.0%61.4%5/70
phi2.7b1.6 GB
77.2%
60.9%6542 ms68.7%55.9%96.3%88.0%60.6%6/70
granite3-moe1.0b821 MB
66.3%
62.5%1044 ms69.4%54.7%99.4%88.1%60.5%7/70
gemma3ne2b5.6 GB
82.5%
63.6%21745 ms70.9%54.7%82.3%89.5%60.3%8/70
llama3.21.0b1.3 GB
77.2%
60.1%3865 ms68.8%55.3%94.1%88.0%60.2%9/70
internlm21.8b1.1 GB
57.9%
66.5%1336 ms73.2%53.4%100.0%88.6%59.9%10/70
falcon-h10.5b448b
77.5%
57.2%17576 ms65.2%57.1%95.0%87.6%59.6%11/70
gemma30.27b291 MB
52.3%
66.6%1109 ms73.3%52.4%100.0%88.9%59.5%12/70
llama2.01.0b1.3 GB
80.5%
65.4%17867 ms72.3%54.1%74.0%90.7%59.4%13/70
tinyllama1.1b637 MB
68.0%
63.0%1243 ms70.4%54.8%100.0%77.1%59.2%14/70
yi-coder1.5b866 MB
69.8%
61.5%2167 ms69.3%55.3%96.3%81.7%58.9%15/70
gemma22.0b1.6 GB
78.6%
51.0%6095 ms60.9%59.7%100.0%88.5%58.7%16/70
ternary-bonsai1.7b3.4 GB
91.4%
53.7%13473 ms62.9%58.3%81.0%86.2%58.7%17/70
phi3:latest3.8b2.2GB
78.9%
56.8%44801 ms64.8%57.3%97.5%81.3%58.7%18/70
vertalily1.21.0b600 MB
67.4%
67.1%1880 ms73.7%53.6%82.0%86.4%58.6%19/70
smollm1.7b990 MB
83.9%
44.8%4644 ms55.8%61.4%100.0%90.6%58.5%20/70
bonsai1.7b248 MB
76.7%
61.0%3672 ms68.8%55.5%79.0%89.4%58.4%21/70
qwen30.6b522 MB
74.2%
55.5%5306 ms64.2%57.4%95.7%87.5%58.3%22/70
gemma3-it1.0b689 MB
65.4%
65.1%8975 ms72.1%53.9%88.7%88.6%58.3%23/70
granite3.1-moe1.0b1.4 GB
78.9%
63.0%5409 ms70.4%55.0%73.5%88.3%58.2%24/70
qwen2-math:1.5b1.5b935MB
95.0%
36.2%15537 ms48.9%68.2%100.0%92.4%58.2%25/70
qwen2.50.5b397 MB
71.9%
57.6%2590 ms66.0%56.9%95.9%83.5%58.1%26/70
medpsy1.7b4.1 GB
90.2%
49.3%28452 ms59.4%60.9%81.0%92.5%58.0%27/70
atem-wisdom1.5b986 MB
69.6%
62.1%7373 ms69.7%55.7%96.3%78.7%57.9%28/70
deepscaler1.5b3.6 GB
100.0%
33.1%19449 ms46.4%65.1%100.0%88.3%57.7%29/70
falcon31.0b1.8 GB
73.2%
57.9%4049 ms65.1%56.7%96.3%79.7%57.7%30/70
lfm20.35b229 MB
79.7%
44.9%1370 ms55.9%61.3%98.5%87.2%57.7%31/70
smollm21.7b1.8 GB
66.0%
65.6%2845 ms72.5%54.1%78.0%87.8%57.3%32/70
stablelm-zephyr3.0b1.6 GB
66.7%
58.0%2475 ms66.3%56.5%98.1%80.4%57.3%33/70
granite40.35b708 MB
67.5%
56.5%1956 ms64.6%57.0%97.3%81.4%56.9%34/70
ministral-33.0b3.0 GB
67.4%
59.3%9955 ms67.5%56.2%86.4%89.0%56.7%35/70
hunyuan-instruct1.8b1.1 GB
98.3%
33.4%15320 ms46.7%65.3%100.0%82.1%56.6%36/70
deepseek-coder1.3b776 MB
57.2%
63.3%2095 ms70.7%54.5%96.3%77.3%56.2%37/70
granite3.22.0b1.5 GB
83.9%
50.8%8569 ms60.7%59.6%81.0%84.3%56.2%38/70
gemma3270m291MB
61.0%
56.1%7262 ms64.6%56.6%96.3%87.5%56.2%39/70
hermes33.0b2.0 GB
56.7%
60.9%4374 ms68.7%55.7%91.0%86.5%55.6%40/70
qwen2.5-coder-instruct3.0b2.1 GB
48.2%
66.1%3428 ms72.8%53.0%96.3%77.1%55.1%41/70
deepseek-r11.5b1.1 GB
62.4%
53.4%10296 ms63.7%58.1%96.9%78.6%54.1%42/70
smollm2:360m360m726MB
57.4%
58.9%7121 ms67.1%56.2%96.2%75.1%54.1%43/70
onellm-doey-v11.0b1.3 GB
47.2%
67.9%972 ms74.3%53.0%96.2%63.4%54.0%44/70
interlm2.51.8b3.8 GB
63.0%
58.9%7883 ms67.1%56.0%100.0%59.3%53.4%45/70
lfm2-700m700m468MB
60.1%
47.0%14270 ms55.6%61.1%100.0%87.4%53.2%46/70
gemma:2b2b1.7GB
59.6%
45.9%15646 ms54.7%60.8%99.0%75.9%52.6%47/70
smollm2360m725MB
59.6%
53.1%8009 ms61.6%58.5%97.1%72.0%52.5%48/70
smollm2135m270MB
61.6%
52.3%134129 ms61.3%58.6%93.5%72.1%52.2%49/70
smollm360m229MB
64.0%
41.0%40120 ms51.8%62.5%95.5%85.4%51.5%50/70
stablelm21.6b982 MB
54.3%
45.1%2171 ms54.1%61.6%100.0%81.1%51.3%51/70
qwen20.5b352 MB
56.0%
51.6%6729 ms60.7%58.9%93.8%74.3%51.3%52/70
qwen2.5-instruct1.54b1.1 GB
56.1%
50.0%3478 ms58.0%59.9%94.0%74.5%50.9%53/70
llama3.2-instruct1.0b807 MB
47.2%
45.8%3746 ms54.7%61.4%97.0%85.9%50.1%54/70
h2o-danube30.5b318 MB
51.2%
44.3%10128 ms53.5%61.6%100.0%81.7%49.9%55/70
deepseek-r1-distill-qwen1.5b1.1 GB
69.6%
39.3%9924 ms51.4%63.7%100.0%64.0%49.8%56/70
TinyDolphin1.1b636MB
54.0%
46.6%9229 ms56.3%60.9%100.0%70.7%49.6%57/70
tinyllama:latest1.1b637MB
56.7%
46.0%8466 ms51.8%61.2%100.0%75.3%49.2%58/70
smollm135m91MB
56.8%
38.6%12748 ms49.9%63.5%100.0%78.2%49.1%59/70
olmo1b791 MB
46.3%
42.9%15029 ms52.4%62.5%100.0%83.5%48.7%60/70
qwen2.5-coder0.5b397MB
49.0%
46.9%7207 ms57.2%60.4%94.6%72.8%48.4%61/70
phi-4-mini-instruct4.0b2.5 GB
57.4%
48.5%70452 ms57.5%59.2%82.0%69.2%48.1%62/70
qwen1.50.5b394 MB
33.1%
47.9%723 ms56.3%60.4%97.0%75.3%47.4%63/70
supra-50m-instruct0.0518b104 MB
37.3%
25.1%581 ms25.1%68.2%100.0%44.3%35.8%64/70
phi-33.8b2.2 GB
27.3%
20.0%9997 ms36.0%70.0%95.3%51.2%33.3%65/70
bloom560m539MB
35.8%
24.5%20277 ms38.9%68.8%97.0%24.8%32.4%66/70
bloom-560m0.8b1.6 GB
37.7%
22.7%7540 ms23.1%66.9%100.0%17.1%30.5%67/70
asena-esp320.121b2.7 MB
19.7%
6.1%1677 ms24.9%74.5%100.0%47.2%28.3%68/70
willow-alpha0.3b176 MB
12.4%
11.9%303 ms29.6%71.7%100.0%31.3%28.2%69/70
minicpm51.0b688 MB
0.0%
5.2%409 ms12.2%71.3%100.0%45.6%25.2%70/70
iShowing 70 evaluated records from OpenVals Performance & Trust validation suite.
VIEW METHODOLOGY

Research, books, media, and IEEE publications behind the trust layer.

View founder profile
14

IEEE-indexed AI and ML publications

2

Authored books across cyber and law

Media

Forbes, TradeFlock, Elets, and more

Forbes
TradeFlock
Elets Healthcare
Blindwink
Pride India Awards
The CConnects
And more.....
Engagement Packages

Start with the right level of AI assurance.

Full validation engagement

Enterprise AI Validation

A deeper evaluation of model quality, safety, reliability, bias, and deployment readiness.

Scoped program
Adversarial testing

AI Security

Manual and automated pressure testing for jailbreaks, prompt injection, leakage, and abuse cases.

Attack simulation
Audit preparation

Compliance Readiness

Evidence, controls, and governance artifacts for teams preparing for AI oversight.

Regulatory support

Book Your AI Trust

Get a fast baseline of your model's reliability, safety, compliance, and deployment risk before you invest in a full validation program.