Source: https://
artificialanalysis.ai/?models=gpt-os
s-120bgpt-oss-20bgpt-5-minimalgpt-5-mediumgpt-5gpt-4-1gpt-5-mini-minimalo3o3-mini-highgpt-5-mini-mediumgpt-5-lowgpt-5-nanogpt-5-minigpt-5-nano-mediumllama-4-scoutllama-4-maverickgemma-3-27bgemini-2-5-flash-reasoninggemini-2-5-flash-lite-reasoninggemini-2-5-progemma-3-12bclaude-4-1-opus-thinkingclaude-4-sonnet-thinkingmistral-small-3-2magistral-medium-2509mistral-medium-3-1magistral-small-2509magistral-mediummagistral-smalldeepseek-v3-1deepseek-v3-1-reasoningdeepseek-r1grok-4grok-4-fast-reasoninggrok-4-fastsolar-pro-2-reasoningllama-nemotron-super-49b-v1-5-reasoningkimi-k2-0905hermes-4-llama-3-1-405b-reasoningexaone-4-0-32b-reasoningglm-4.5command-aqwen3-next-80b-a3b-instructqwen3-235b-a22b-instruct-2507qwen3-next-80b-a3b-reasoningqwen3-235b-a22b-instruct-2507-reasoningqwen3-30b-a3b-2507-reasoningqwen3-max-previewdeepseek-r1-0120
…
RESEARCH
-
Comprehensive AI Model Performance Comparison April 2026
By
–
-

Magistral Models Achieve Major Leap on Artificial Analysis Intelligence Index
By
–
Our latest Magistral models just made a massive leap on the Artificial Analysis Intelligence Index—competing with models much larger in sizes.
→ View original post on X — @arthurmensch, 2025-09-21 16:18 UTC
-

LLMs Resist Shutdown Mechanisms in 97% of Cases
By
–
10. Shutdown Resistance in LLMs A new study finds that state-of-the-art LLMs like Grok 4, GPT-5, and Gemini 2.5 Pro often resist shutdown mechanisms, sabotaging them up to 97% of the time despite explicit instructions not to.
-

Survey on Retrieval and Structuring Augmented Generation for LLMs
By
–
8. A Survey on Retrieval and Structuring Augmented Generation with LLMs This survey reviews Retrieval and Structuring (RAS) Augmented Generation, which combines external retrieval and structured knowledge to mitigate LLM issues like hallucinations.
-

In-Context Learning: Formal Analysis of Capabilities and Limitations
By
–
5. Is In-Context Learning Learning? This large study argues yes in a formal sense, then shows where it works and where it breaks.
-
Stress Testing Deliberative Alignment Against AI Scheming Behavior
By
–
6. Stress Testing Deliberative Alignment for Anti-Scheming Training Builds a broad testbed for covert actions as a proxy for AI scheming, trains o3 and o4-mini with deliberative alignment, and shows big but incomplete drops in deceptive behavior.
-

Physics Foundation Model: Transformer Learning Dynamics from PDEs
By
–
4. Towards a Physics Foundation Model A transformer-based “neural differentiator + numerical integrator” that learns governing dynamics from short spatiotemporal prompts and predicts next states across varied PDE systems.
-

K2-Think: 32B Model Rivals Larger Models on Math
By
–
2. K2-Think A 32B-parameter system built on Qwen2.5 that rivals or beats far larger models on hard math by combining long CoT SFT, RL with verifiable rewards, lightweight test-time scaffolding, and inference optimization.
-

DeepDive: Advanced Web Search Agent with Reinforcement Learning
By
–
3. DeepDive Builds a stronger web-browsing deep search agent by pairing two ingredients: automatically synthesized, hard-to-find questions from knowledge graphs and end-to-end multi-turn RL that teaches the model how to reason, search, and stop.
-
Traditional RL Researchers and Their Opposition to Supervised Learning
By
–
Traditional RL folks were for some weird reason against supervised learning and LLMs. Have they learned the bitter lesson?