AI news5
Runway opens Praxis-1 weights under Apache 2.0
Runway released Praxis-1's weights and fine-tuning code on 11 November, extending access beyond its September launch partners. The 27-billion-parameter robotics model predicts video and actions together, with Runway reporting 46.2% success across 1,200 RoboLab-120 benchmark episodes.
runway.com ↗Why it matters
Opening the model after partner deployments lets robotics teams inspect and adapt the same foundation for their own hardware.
Google adds a US$0.10-per-million-token decision API
Google's 11 November Gemini Decisions API scores up to 1,000 questions against a shared text, image, audio, video or PDF input. It charges US$0.10 per million input tokens and no output-token fee, with Google reporting 150–250 millisecond end-to-end latency.
blog.google ↗Why it matters
The new scoring endpoint adds a routing and policy-checking layer alongside the generative model releases covered earlier this week.
Neo4j releases an embedded graph database for AI agents
Neo4j launched Relate on 11 November as an open-source graph database that runs inside an application, command-line tool or browser. Its launch demonstration uses Claude Code to audit the 24 direct dependencies of Google's Agent Development Kit through separate temporary graphs.
neo4j.com ↗Why it matters
Keeping relationships and similarity searches in one local database gives agent builders a simpler way to preserve connected context.
MiniMax H3-Image adds 4K generation and 16-image references
MiniMax's 11 November English announcement introduces H3-Image for image generation and editing, with support for up to 16 reference images. The company says it can produce a 4096-by-4096 image in under 10 seconds and offers access through its API, MiniMax Code and Hailuo AI.
minimax.io ↗Why it matters
The still-image release expands creative workflows beyond the previously covered MiniMax language model, with reference-based edits as well as generation.
UK public funds back Fractile's £500-million-plus round
The UK government confirmed on 11 November that Sovereign AI and the British Business Bank joined Fractile's Series C, worth more than £500 million. The round values the London and Bristol AI-chip company at over £5 billion, with CapitalG, Lightspeed and Redpoint also named as investors.
gov.uk ↗Why it matters
Public capital is backing domestic chip design as governments seek a larger stake in AI hardware and its supply chains.
AI in the nonprofit sector4
CareSource opens a US$1 million responsible healthcare AI call
The CareSource Foundation's 10 November call offers up to US$500,000 a year for two years for an implementation-ready healthcare AI solution. A US 501(c)(3) nonprofit in good standing for at least three years must lead or fiscally sponsor the proposal, with concepts due at 3:59 pm AEDT on 7 January 2027.
caresource.com ↗Why it matters
The call gives nonprofit-led partnerships a funded route to test care improvements with explicit privacy, bias, validation and human-oversight requirements.
Thomson seeks responsible-AI mentors for East African media
Thomson Media's AGILE Africa call includes responsible AI and product governance among six training and mentoring specialisms for media organisations in Kenya, Tanzania and Uganda. Applicants can be based anywhere but need extensive relevant East African experience, and must propose a daily rate in euros before 7:59 am AEDT on 19 November.
thomsonfoundation.org ↗Why it matters
The paid-assignment route connects AI governance expertise with independent media's operating needs rather than offering a general technology grant.
Apart's online AI research sprint starts today with US$2,000 in prizes
Apart Research's 13–15 November AI x Epistemics sprint invites teams of one to five to build evaluations, verification tools or decision-support research. The five cash prizes total US$2,000, with submissions due at 10:59 pm AEDT on 16 November and selected teams invited to seek continued fellowship support.
apartresearch.com ↗Why it matters
Today's start opens a short, practical route for researchers and public-interest builders to test whether AI advice and evidence can be trusted.
Helmholtz's collaborative AI call closes at 9:59 am AEDT today
Helmholtz AI's 2026 project call closes at 9:59 am AEDT on 13 November, equivalent to 12 November at 11:59 pm CET. At least two Helmholtz centres must apply together for projects of up to three years, with up to €250,000 from the association's fund and at least matching centre financing.
helmholtz.ai ↗Why it matters
The deadline-day reminder is useful for already assembled research collaborations, while external nonprofits can participate as partners rather than apply independently.