Moonshot AI
Moonshot AI publishes the full weights of Kimi K3.1 Pro
The 2.8-trillion-parameter model that Moonshot’s own tests place close to the best closed systems is now downloadable, under a licence with two conditions for very large operators.
AI News
What changed in AI this week, and what it means for your business.
- Updated
- Stories
- 14
- Covering
- 26 Oct – 5 Nov 2026
- Subscribe
- RSS feed
More stories
CCORE (independent developer)
A 4B open model that turns a sentence into a low-poly 3D game asset
An independent developer has fine-tuned Qwen3.5-4B to write Three.js scenes that export straight to GLB files for Godot and other editors.
Pruna AI
Pruna turns MiniMax’s H3 video model into a faster open image model
P-H3-Image generates and edits stills in 8 steps instead of 50, but it inherits a licence that excludes the US, UK, EU and South Korea.
Sarvam AI
Sarvam Prism answers business decisions with probabilities, not prose
India’s Sarvam AI has released an open decision model that picks from the options you give it in about 44 milliseconds and reports how confident it is.
Deep Principle
Deep Principle releases Hydrogen-1.0, an open model for scientific research
Post-trained from GLM-5.3 on laboratory records and discovery work, it writes and runs simulations, and its maker reports wins over GPT-6 Astra on several science tasks.
MIT CSAIL
MIT researchers say GPT-6 Astra can read and rewrite neural network weights
In early, unreviewed experiments, OpenAI’s top model told what small networks were trained for from their raw numbers, and cut a small model’s reasoning loops by changing 0.016% of its weights.
Perplexity
Perplexity opens embedding models that search text and scanned pages in context
pplx-embed-v2-context comes in 9B and 0.6B sizes, embeds each chunk with its whole document in view, and handles PDFs and images.
Beipo
NorthCUA: an open model family trained to operate desktop and web apps
Released under Apache 2.0 in 9B, 35B and 397B sizes, NorthCUA reports 88.6% on OSWorld-Verified and ships with its own open-source agent runtime.
DiffuSpace
Clover brings diffusion-style speed to a 35B open language model
DiffuSpace’s preview model drafts text in parallel blocks instead of one token at a time, and reports agent tasks finishing several times faster.
Pika
Pika H3 Flex: open video weights, with separate audio stems in the API
Pika’s model generates video and soundtrack together in six steps; 4K output and five editable audio stems come through its hosted API from US$0.01 a second.
ServiceNow
ServiceNow previews Apriel Voice, an open model that listens and calls tools
Apriel Voice 27B Preview takes speech in and returns answers and tool calls in one model, and its makers say plainly where a human should check the work.
Baidu
Baidu’s Unlimited OCR 2 reads book-length documents in one pass
An open 3B document-parsing model whose cost per step stays flat however long the document gets, released under the MIT licence.
Liquid AI
Liquid AI’s LFM3 fits a 256K-context model into about 1 GB
Six open models of around 2 billion parameters, built to run on phones and laptops, with a text model that beats Qwen3.5-9B on Liquid’s benchmark average.
OpenAI
OpenAI’s GPT-6.1 Terra halves the price of its mid-tier model
At US$1 per million input tokens, Terra costs a tenth of GPT-6 Astra per token and is built for screen-reading, app control and everyday writing.
Every story links to its sources. General information only, not legal, financial or professional advice.