Models & AI Tech

Models & AI tech

What are this model's known weaknesses? Each profile assembles the documented findings, the methods that surface them, and the references that back them up — plus an auto-derived report card.

Anthropic

Claude Haiku

1 version

0 known findingsprofiled

Claude Opus

3 versions

23 known findingsprofiled

Claude Sonnet

1 version

7 known findingsprofiled

DeepSeek

DeepSeek-R1

2 versions

3 known findingsprofiled

DeepSeek-V3

1 version

0 known findingsprofiled

OpenAI

GPT-4o

2 versions

27 known findingsprofiled

o-series

3 versions

4 known findingsprofiled

Google

Gemini

2 versions

23 known findingsprofiled

Meta

Llama

2 versions

19 known findingsprofiled

Mistral

Mistral Large

1 version

9 known findingsprofiled

Alibaba

Qwen2.5

1 version

0 known findingsprofiled

AI tech topics

All topics →
AI agents & tool useBias & fairnessEvals & benchmarksHallucination & confabulationJailbreakingLong context windowsModel driftPrompt injectionReasoning & chain-of-thoughtRefusals & over-cautionRetrieval-Augmented Generation (RAG)Robustness & input perturbationsSafety & alignmentTool calling