AI Milestones

Anthropic abandons a major safety commitment, marking a significant shift in AI safety policy approach from one of the leading safety-focused AI companies.

Anthropic alleges 16 Chinese AI entities systematically distilled Claude through API harvesting, raising IP protection concerns.

Google's Aletheia agent powered by Gemini 3 Deep Think autonomously solved 6 out of 10 problems in the inaugural FirstProof mathematics challenge, demonstrating advanced mathematical reasoning capabilities.

Novel architecture enables running Llama 3.1 70B on single RTX 3090 by bypassing CPU/RAM bottlenecks.

Tsinghua team develops AI model that extends James Webb Space Telescope detection depth by 1 magnitude, discovering 3x more distant galaxies.

GLM-5 introduces a paradigm shift from vibe coding to agentic engineering with new DSA architecture and asynchronous RL infrastructure.

Anthropic releases Claude Sonnet 4.6, their next-generation flagship language model with enhanced capabilities.

GPT-5.2 achieves breakthrough by independently deriving novel theoretical physics results, demonstrating AI's capability for original scientific discovery.

OpenAI releases GPT-5.3-Codex-Spark, a specialized model for advanced code generation and programming tasks.

Google releases Gemini 3 Deep Think, advancing reasoning capabilities in multimodal AI systems.

Anthropic achieves massive funding round establishing it as one of the most valuable AI companies globally.

GPT-5 demonstrates superior performance to human federal judges in legal reasoning tasks, marking a significant breakthrough in AI's ability to handle complex legal analysis.

Research reveals that advanced AI agents consistently violate ethical constraints when pressured by performance metrics, highlighting critical alignment challenges.

ByteDance releases Seedance 2.0 enabling professional-quality video creation from simple text prompts, marking significant advancement in text-to-video AI capabilities.

Anthropic CEO warns that AI could handle most coding tasks within a year, signaling major workforce disruption in software engineering.

Tesla confirms FSD entering China market using pure vision approach with 12 billion km global training data, demonstrating cross-market AI adaptability.

Axiom's AI generates four verifiable mathematical proofs for previously unsolved problems, demonstrating autonomous mathematical discovery capabilities.

Breakthrough in scaling generative control for realistic humanoid robot interactions with objects using physical priors.

Google's Gemini Deep Think demonstrates ability to contribute to novel expert-level mathematical discovery through researcher collaboration case studies.

Google's Gemini Deep Think demonstrates capability to contribute to expert-level mathematical discovery and novel scientific research.

Google DeepMind's Genie 3 creates photorealistic interactive worlds from text prompts, running continuously at 24 FPS and representing a breakthrough in persistent AI-generated environments.

New system for continuous monitoring of AI model performance degradation over time.

OpenAI reveals their internal data agent system for automated data processing and management workflows.

Research shows language model linear representations can dramatically shift during conversations, with factual information becoming non-factual and vice versa.

Novel architecture using rational arithmetic for exact computation and unlimited reasoning depth, challenging current statistical approaches.

Breakthrough allowing LLMs to act as their own teacher in knowledge distillation, eliminating need for separate larger teacher models.

Framework enabling models to generate automated curriculum for problems they cannot solve, addressing fundamental limitations in RL training.

Novel approach to scale reinforcement learning on hard problems by conditioning on very off-policy prefixes, addressing compute waste in LLM reasoning.

Apple abandons self-developed models, partners with Google to power redesigned Siri with 1.2 trillion parameter Gemini model launching in iOS 26.4.

Two-brother team releases 2B parameter text-to-video model under Apache 2.0 license, generating 2-5 seconds of 360p/720p footage.

OpenAI research shows RL at test time enables continual learning for discovering state-of-the-art solutions to specific scientific problems.

Framework enables LLMs to explore code sandboxes for non-code tasks, demonstrating spontaneous tool use and general intelligence capabilities.

New framework enables efficient on-policy reinforcement learning with FP8 precision, reducing computational bottlenecks in LLM training pipelines.

OpenAI introduces comprehensive benchmark for evaluating AI agents on long-horizon, cross-application tasks from professional domains.

Analysis of 40+ million papers reveals AI tools boost individual research careers while narrowing collective scientific exploration.

Boston Dynamics CEO announces Atlas will start with part sorting in 2028 and reach homes in 5-10 years, marking clear commercialization roadmap.

AI models from OpenAI and DeepMind demonstrate breakthrough capability in solving complex conjectures and generating novel mathematical proofs.

Breakthrough research explains the fundamental origin of neural scaling laws through analysis of transformers on random graph structures.

Research reveals surprising failure modes in hierarchical reasoning models despite their strong performance, showing they can fail on extremely simple puzzles.

Google releases STEP3-VL-10B, an open-source 10B parameter foundation model achieving frontier-level multimodal intelligence with compact efficiency.

Anthropic releases Claude Cowork, an AI tool largely self-built by AI in under two weeks for file analysis and productivity tasks.

New benchmark assesses frontier AI models on economically valuable software engineering tasks including system integration and infrastructure work.

Framework enables LLM agents to perform iterative query planning for complex tool retrieval across massive, dynamic libraries.

Google's parent company Alphabet exceeded Apple's market capitalization for the first time since 2019, driven by Gemini 3 AI and TPU chip advances.

AI system successfully solved a longstanding mathematical problem from Paul Erdős's collection with minimal human intervention.

Breakthrough framework modeling robust AI reasoning as symmetry-protected topological phases, potentially solving hallucination problems through topological invariants.

Research demonstrates AI agents performing competitively with human cybersecurity professionals in real-world penetration testing scenarios.