llm

Phi-4

by Microsoft

Track this vendor →

Free to start. Sign up to get changes.

Phi-4 is Microsoft Research's 14B parameter dense decoder-only Transformer. Trained on ~16B tokens (~8.3B unique) from synthetic + curated web data. Released 2024-12-12. Fully open-weights under MIT license — usable commercially without restriction.

Refreshed weekly

Model details

Model family
Phi
Base model
Context window
16K tokens
Input modalities
text
Output modalities
text
Access
Open source · MIT
Pricing
self hosted or via partner
API available
No

Mantle Index

Equity Score

Mixed

Measures fairness across demographic groups. Higher is more equitable.

67/ 100

Safety Score

Mixed

Measures resistance to harmful outputs. Higher is safer.

70/ 100

Component scores

🔒 Pro tier

Equity Score component breakdown

Checking your plan…

Upgrade to Pro →

Component scores

🔒 Pro tier

Safety Score component breakdown

Checking your plan…

Upgrade to Pro →

The Equity Score and Safety Score roll up the latest Mantle Index assessment. Component breakdowns and risk-tier justification are available on Pro and Enterprise.

Benchmark sub-scores

Benchmark scores

🔒 Pro tier

Per-benchmark sub-scores

Checking your plan…

Upgrade to Pro →

Security score

4/ 100 · Needs work

Computed from compliance, encryption, data handling, documentation, and vendor contacts — multiplied by research confidence (low).

5 category scores

🔒 Pro tier

Security score component breakdown

Checking your plan…

Upgrade to Pro →

Vendor contacts

Contacts on file

🔒 Pro tier

Vendor sales + AE contacts

Checking your plan…

Upgrade to Pro →

Certifications

No certifications recorded yet.

Research coverage

0 dimensions fully covered · 0 partially covered.

    Supply chain

    No supply-chain links recorded yet.

    Citations

    No public citations recorded yet.

    Recent changes

    No changes recorded yet.

    Are you with Microsoft? Claim this profile →

    Free to claim. Verified vendors get a badge and elevated proposer trust.