Cognition raised $2 billion in funding for its coding assistant Devin at a $48 billion valuation. The company said its annualized revenue run-rate grew from $492 million to $900 million since its May round at a $26 billion valuation. Investors’ backing and the new funding should let Cognition expand and train its own model to reduce reliance on costly third-party AI models and target breakeven.
Magic, the AI startup founded by Eric Steinberger and Sebastian De Ro, published a research update claiming its language-model pretraining is over 10x more compute-efficient than recipes behind leading open-weight base models. It says matching DeepSeek V4 Pro’s pretraining took 50x less compute, costing about $0.5M on Nvidia GB200 (and scaling to about $4M). The result is a new cost-and-benchmark narrative based on its own held-out bits-per-byte loss measurements and a next stated focus on long-horizon reinforcement learning and deployment safety policies, while noting the figures are unverified and no model has shipped.
Pathway developed its brain-inspired BDH model architecture and trained its BDH-CQ reasoning system on Amazon SageMaker HyperPod. BDH-CQ reached 29.5% pass@2 on ARC-AGI-1 at US$0.0007 per task. The work shifts reasoning from token-by-token text traces to iterative latent computation with fewer active neurons, aiming to improve cost efficiency and scalability in production training and inference.
Pathway developed its brain-inspired BDH (Dragon Hatchling) architecture and ran training on Amazon SageMaker HyperPod for scaling out distributed GPU compute. BDH-CQ reached 29.2% pass@2 on the ARC-AGI-1 benchmark at a stated cost of US $0.0007 per task. The shift updates model reasoning in a recurrent latent state without generating long chain-of-thought traces, changing the cost–accuracy tradeoff and how memory and computation are handled at inference.
The talk “From S3 to GPU in One Copy” explains how SpiralDB’s Vortex columnar format enables training data to be streamed from S3 into GPUs with projection and filter pruning instead of decompressing/reprocessing everything. It reports S3-to-GPU scans at around 30 times faster than Parquet. As a result, only selected columns/segments are read and decompressed for each iteration, reducing movement and decision overhead during ML training.
The Sequence Knowledge 928 post argues that model distillation falls short of a simple “95% of a 70B teacher” promise for a 7B student because the remaining performance gap can show up in specific behaviors like calibration, recovery after tool failures, and judgment. The article anchors the discussion on the “missing 5%” between a 7-billion-parameter student and a 70-billion-parameter teacher. As a result, it frames distillation as harder than expected and suggests the gap may be concentrated in critical capabilities rather than spread uniformly.
Danijar Hafner left Google DeepMind to start a new stealth-mode robotics startup focused on AI agents that can plan ahead for unfamiliar situations. He is building on Dreamer 4, which learned diamond mining from an offline dataset of recorded gameplay videos without directly interacting with the game. The work shifts from game-based world models to physical humanoid robots designed to handle new indoor layouts and respond to unexpected events in real homes.
Mistral AI announced a €3 billion Series D funding round at a post-money valuation of over €21 billion led by Samsung Electronics. The company plans to use the capital to expand its compute capacity for training larger models and to support international growth. The funding underwrites its shift from a pure model maker into a European neocloud with tools like regional processing and pre-financed data-center build-out.
Every AI story that matters,
in your inbox by 8am.
TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the
day in two minutes. Follow companies and topics for alerts, or get the
briefing in Slack. Free, no spam, unsubscribe anytime.
Reading TLDRocket needs no cookies, and the readership counts we rely on come from
our own cookieless analytics. Google Analytics is the exception: it sets cookies and
reports to Google, so it stays switched off until you allow it. You can change your
mind any time from “Cookie settings” in the footer.
Strictly necessary
Session security and form protection (tldrocket-session,
XSRF-TOKEN, 2 hours). The site cannot work without them,
so they need no consent.
Always on
Google Analytics 4 (_ga,
_ga_<id>, up to 2 years). Measures which
stories and sections readers use. Google acts as a third-party processor and may
store the data outside the EU. No advertising, no profiling, no data sold.
