Fireworks AI closed a $1.505 billion Series D led by Atreides Management, Index Ventures, and TCV, pushing its valuation to $17.5 billion. The company surpassed $1B ARR running specialized inference on open models, serving customers including major AI labs on NVIDIA Blackwell infrastructure.
Former DeepMind researcher Andrew Dai's visual AI startup Elorian raised a $55 million seed round at a $300 million pre-product valuation, led by NVIDIA and Menlo Ventures. The unusual valuation and strategic investor lineup signal intense demand for visual AGI capabilities.
LM Studio released Bionic, an AI agent designed specifically for open-weight models with voice input, local voice transcription, flexible execution across local hardware, LM Link, or LM Studio Secure Cloud, and a companion iPhone app.
Reinforcement learning with verifiable rewards — without human-annotated data — is scaled to a trillion-parameter model, eliciting emergent chain-of-thought reasoning at unprecedented scale.
A method to scale the unrolled depth of looped transformers without increasing stored parameters, analyzing the residual-scaling problem unique to tied-parameter architectures.
A frozen small LM is made more capable and cheaper by grafting byte-exact KV-cache states from verified knowledge, preserving bit-exact logit reproduction under deterministic inference.
A unified framework for discrete denoising diffusion models offering parallel generation and iterative global refinement as an alternative to autoregressive modeling for discrete data.
Length-penalized RL can shorten CoT reasoning while hiding misleading hints that still steer the model's answer, raising safety concerns about CoT monitoring under efficiency constraints.
An AI coding agent that generates machine-checkable proofs alongside code, combining LLM code generation with formal verification to guarantee correctness.
An open-source explainable inference gate that detects and blocks prompt injection attacks on LLMs, providing a lightweight security layer for AI applications.
Debug agents that automatically diagnose and fix failing Playwright test scripts in CI/CD pipelines, reducing manual test maintenance for frontend teams.
University of Illinois released an AI teaching assistant platform that helps instructors scale personalized feedback and tutoring using LLMs, built for educational contexts.
A QA automation agent that analyzes your application's source code to understand UI structure, then writes and executes end-to-end tests more intelligently.
Fora closed a $60 million Series D at a $1B valuation, led by Forerunner and Tactile Ventures. The AI-augmented travel advisor platform now serves 15,000+ active advisors with $3B+ in booked travel.
Munich-based Microagi raised a $55 million seed round (largest in German startup history) led by Hummingbird, building data infrastructure for physical AI and robot deployment.
Sable raised $45 million led by Sequoia Capital and 8VC for its agent platform 'Aidan' that combines computer use, vision, and voice for live customer experiences. Customers include Notion and Decagon.
Singapore-based Whale raised a $40 million Series C3 extension ($100M total Series C), led by CMB International and SMBC Asia Rising Fund, serving 800+ enterprise customers across Asia-Pacific and North America.
Bunkerhill Health raised $25 million Series B led by Khosla Ventures for its Carebricks platform that lets hospitals build AI agents for imaging, registry management, prior authorizations, and triage.
Source: