Odyssey, an AI lab specializing in world models, announced on June 17, 2026 that it had raised $310 million in a Series B round. The post-money valuation reached $1.45 billion, and the company is shifting toward using AWS Trainium chips through a new partnership.
April 2026 · AI Frontier Report
Google Reboots Its Coding "Strike Team" to Chase Anthropic
With Sergey Brin and Koray Kavukcuoglu involved, Google is pushing Gemini into "midtraining" — the stage between pretraining and post-training — to crack long, complex coding tasks. The wider story: AI is pivoting from training toward inference and the real world.
~100%
Claude's share of internal code generation in one example
~50%
Gemini's share in the same example
$310M
Odyssey Series B for world simulation
Code generation, one internal example
Why Anthropic's Claude is seen as the coding leader
Despite ample TPUs and Gemini, Google lags in coding-specialized training loops — the case for "midtraining."
Talent moves that triggered the reshuffle
Noam Shazeer
Departed → OpenAI
John Jumper
Nobel laureate → Anthropic
The same pivot, three other signals
AI moves from training toward inference & the real world
Memory · Micron
Quarterly revenue projected at ~$50B ; guidance above Wall Street as DRAM/NAND demand spreads from training to inference.
Robotics · a "ChatGPT moment"?
Foundation & world models (NVIDIA Cosmos, π) advance, but a real-world data shortage remains — some say we are still only at "GPT-2.5" level .
World Models · Odyssey
$310M Series B at a $1.45B post-money valuation, led by Natural Capital, on AWS Trainium — simulating the physical world for robotics, gaming and science.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…