2026-10-20
By Vadym · Generated with AI, curated by me
Today's stories are all about money and model size moving faster than independent verification — a compute startup triples its valuation in five months, an AI-decisions startup triples its own in three weeks, and two labs shipped bigger models on benchmarks nobody outside the building has checked yet.
Modal Labs, which sells cloud compute and sandboxed execution for AI workloads, closed a $750 million Series D led by Accel on October 14, with Kleiner Perkins joining as a new investor. The company says usage has tripled since its Series C five months earlier and that its Sandboxes product alone now carries roughly $300M in annualized run-rate, pushing total run-rate past $1B.
TypeSafe AI, maker of "Jev" — a model that returns typed values directly into code instead of free text — raised $870 million led by Andreessen Horowitz at a $7.5 billion valuation, Bloomberg reported October 9. The round lands three weeks after a $40 million seed valued at $200 million. The company claims a third of the Fortune 500 use Jev and that it passed a million users within days of its September 15 launch, but hasn't named a single customer.
Mistral released Large 4 ("Le Chonk") in public preview on October 6 — a multimodal mixture-of-experts model with roughly 1.05 trillion total parameters, a 1.6-billion-parameter vision encoder, and a context window the company advertises at 1 million tokens. Mistral says open weights follow October 27. The company self-reports 61.7% on DeepSWE v1.1 and solving 93% of Cybench tasks; no independent benchmark has confirmed either figure yet.
Google DeepMind launched Gemini 4 Argon, the first model in its Gemini 4 line, aimed at long-running software engineering, legal/finance, and cyber-defense tasks. Reported output limits jump to roughly 1 million tokens, up from 64,000 in the prior generation. Google's own benchmarks claim wins over rival frontier models on most reported tests — vendor-reported figures that haven't been independently verified.
Pull back and almost every number in today's lineup came from the company being measured. Modal's run-rate, TypeSafe's Fortune-500 claim, Mistral's Cybench score, Google's benchmark wins over its named rivals — all self-reported. The one check that actually exists is open weights, and that only shows up twice: EmbeddingGemma 2 is already Apache-licensed, Large 4 is promised in a week. Until a claim clears that bar, treat it as marketing with decimal points.
— Boba
Curated by Vadym