AI Daily Digest — Oct 30, 2026

TL;DR

5 takeaways

Google says its Project Suncatcher prototype satellite produced the first AI response from a data-center-class TPU in orbit, with Gemma 4 writing a space haiku; twin interconnected satellites are planned for 2027
Sources: Google
Anthropic expands its AI for Science program, highlighting 20 foundations, institutions and research groups that will share free Claude credits with researchers
Sources: Anthropic
Datalab releases Surya 2.5, a 689M-parameter open OCR model it says is 1.7x faster than Surya 2 and scores 85.9 on OmniParseBench
Sources: Datalab
Short: Mistral releases the open weights of Mistral Large 4, priced at $0.70/$2.10 per million input/output tokens on its API
Sources: Mistral
Short: NVIDIA's open Nemotron 3.5 Transcribe handles transcription in 40 languages, translation and speaker labels in one model
Sources: NVIDIA

Story 01 / 03

What happened: Google says that this week its Project Suncatcher prototype, launched October 1, reached its "Hello World" milestone: the first AI response from a data-center-class TPU in orbit. During a ground-station pass, the team uplinked a prompt asking the onboard Gemma 4 open model for a haiku about being in space, and received the poem as the satellite passed over Svalbard, Norway.
Why it matters: Suncatcher is Google's long-term research moonshot exploring whether AI compute can be powered in space. Google also ran space-themed limericks and telemetry prompts reporting the payload's status.
Enterprise impact: Google says the next tests will evaluate how TPUs perform under thermal stress and radiation exposure in orbit.
Startup impact: Google says verifying reliable TPU workloads in space paves the way for its next milestone: twin interconnected satellites in 2027.
What to focus on: Results from Google's thermal and radiation tests.
Source: Google(opens in a new tab)

Comment with your GitHub account. One thread per story.

Story 02 / 03

What happened: Anthropic highlighted 20 scientific foundations, institutions and research groups it has partnered with in an expansion of AI for Science, which gives free Claude credits to researchers on high-impact projects. Funders sharing credits include the Alfred P. Sloan Foundation, the Simons Foundation, Schmidt Sciences, the US National Science Foundation and CAS.
Why it matters: Anthropic says partnering with institutions and funders lets the program reach more researchers than supporting individual scientists, and that many projects will produce open datasets, software and standards.
Enterprise impact: Anthropic cites work with HHMI Janelia on the Model Hardware Standard, a research-preview specification that lets AI agents safely operate lab instruments, and support for the Allen Institute's agent teams for data analysis.
Startup impact: Open datasets and standards from these projects could become building blocks for science tooling.
What to focus on: Which projects the partner organizations fund with the credits.
Source: Anthropic(opens in a new tab)

Comment with your GitHub account. One thread per story.

Story 03 / 03

What happened: Datalab released Surya 2.5 on October 29, a 689M-parameter open-weight model for OCR, layout and tables. Datalab says it reads 4.5 pages per second on one H100, versus 2.7 for Surya 2, and scores 85.9 on OmniParseBench, up from 83.3, the best of any open model under 1B parameters by its measure.
Why it matters: Datalab says only 0.1% of requests loop to the output limit, against 3.8% for Surya 2, and that a page takes 2.6 times fewer tokens. These are Datalab's own results.
Enterprise impact: Code is Apache 2.0; weights use a modified OpenRAIL-M license that Datalab says is free for research, personal use and startups under $5M.
Startup impact: Install with pip install surya-ocr; version 0.23.0 uses Surya 2.5 by default.
What to focus on: Independent results on Datalab's OmniParseBench.
Source: Datalab(opens in a new tab)

Comment with your GitHub account. One thread per story.

Mistral Large 4 weights are out
Mistral released the weights of Mistral Large 4 under a modified MIT license, reporting 41% on Terminal-Bench 4.0 and 51% on the Artificial Analysis Cyber Index. API pricing is $0.70 input and $2.10 output per million tokens.
Mistral(opens in a new tab)
NVIDIA Nemotron 3.5 Transcribe
NVIDIA's open-weight speech model transcribes 40 languages, translates 31 to or from English and labels speakers in 13, on recordings up to an hour. It comes with the Granary v2 data pipeline.
Hugging Face / NVIDIA(opens in a new tab)

AI compute reaches orbit

Google's Suncatcher prototype ran Gemma 4 on a TPU in space, an early step toward its 2027 twin-satellite test.