News
Mistral Mathstral 2.0 sets new open-source SOTA on MATH and GSM8K-Hard
Mistral Mathstral 2.0 achieves 94.1% on MATH and 98.7% on GSM8K-Hard, new open-source SOTA
Anthropic Claude 4.4 Opus demonstrates 7-day autonomous research agent on novel scientific hypothesis
Claude 4.4 Opus demonstrates 7-day autonomous research agent producing publication-quality chemistry papers
xAI Grok-5 preview adds native causal video generation with physics consistency
xAI Grok-5 preview adds causal video generation with physics consistency for 10-second 720p clips
OpenAI o4-proto-3 achieves closed-loop self-correction on 100+ cycle software debugging tasks
OpenAI o4-proto-3 achieves closed-loop self-correction with 74.2% resolution on 100+ cycle debugging tasks
Mistral Devstral: Specialized 24B Model Optimized for Developer Tool-Chain Integration
Mistral released Devstral (24B), a developer-focused model fine-tuned for seamless integration with IDEs, terminals, git
Meta Llama 4 Maverick: First Open Model to Reach Human-Parity on SWE-Bench Verified
Llama 4 Maverick (open-weight 405B variant) achieves 68.7% resolution rate on SWE-Bench Verified
xAI Grok-4.1 Adds Native Video Understanding and Temporal Action Localization
Grok-4.1 preview integrates dense video captioning, temporal action localization, and event grounding capabilities directly into the core model.
DeepMind AlphaGeometry 2 Solves 92% of IMO Geometry Problems Autonomously
AlphaGeometry 2 reaches 92% solve rate on past 25 years of International Mathematical Olympiad geometry problems
OpenAI o4-proto-2 Demonstrates Self-Evolving Prompt Engineering Loop
OpenAI released an internal preview of o4-proto-2 showing a self-evolving prompt engineering mechanism
Mistral Codestral-Mamba: State-Space Model Achieves SOTA Efficiency on Long-Code Completion
4.1× faster inference than equivalent Transformer models at 256k context, while matching or exceeding pass@1 on HumanEval+, MultiPL-E, and BigCodeBench-hard.
Loading...
There is no more data available~