AI news6
Dream Labs opens a robotics API from $3 an hour
Dream Labs announced its public robotics API and $90 million seed round on 19 October. Dream-1 costs $3 an hour, or $2 for Flash, and targets two-arm robots with parallel grippers. Hardware partner MakerMods says its robot began attempting new tasks within minutes. This is a research beta, with performance varying by hardware and no validation for safety-critical use.
dreamlabs.ai ↗Why it matters
A shared control API could reduce the custom model work required to put inexpensive robot arms to use.
TorchNeuron brings native PyTorch to Trainium in public beta
The 19 October PyTorch announcement scheduled TorchNeuron’s public beta for 20 October; its public repository now provides the backend and installation instructions. It supports eager execution, torch.compile, distributed training and custom NKI kernels on Trainium2 and Trainium3. The code uses the Apache 2.0 licence.
pytorch.org ↗Why it matters
Training teams gain a more familiar software path to alternative AI hardware, while beta compatibility still needs workload testing.
Exa 3 adds $4 search and shorter retrieval context
Exa released version 3.0 of its search API on 19 October, including Lite at $4 per 1,000 requests and Auto at $7. Dynamic Highlights is now generally available at no additional charge. Exa says its cross-result token selection uses 95% fewer tokens than sending full page text; that comparison is the company’s own.
exa.ai ↗Why it matters
Agent builders get a concrete price and context-size tradeoff to test against their existing web retrieval costs.
MiniMax M3.1 Flash launches at $0.15 input and $0.60 output
MiniMax released M3.1 Flash on 19 October through its API, M Plan and MiniMax Code. Prices per million tokens are $0.15 input, $0.60 output and $0.015 cached input. MiniMax reports 70% lower overall agent-task cost than its predecessor, rather than a 70% cut to every token price; model weights are promised later.
minimax.io ↗Why it matters
This standalone model release, beyond earlier Bedrock listings, gives teams a new low-cost option for testing complete agent workflows.
Mercury 3 pairs a million-token context with launch pricing
Inception launched its Mercury 3 diffusion language model on 19 October with a one-million-token context window. Partner TinyFish reports completing 50% of 50 live website navigation tasks in its production harness; this is a partner-reported test. The model is available through Inception and other API providers, with launch pricing of $0.10 input and $0.35 output per million tokens, half the stated regular rates.
inceptionlabs.ai ↗Why it matters
The new base model expands beyond the earlier Mercury Voice product, offering another latency and cost option for interactive agents.
Notion sets two monthly AI credit allowances
Notion’s 19 October update combines a rebuilt agent with a new Business and Enterprise credit system. Each member normally gets 1,500 Standard and 500 Select credits monthly; the separate allowances do not roll over, and affected features pause when their allowance runs out unless an admin permits additional workspace credits. New workspaces changed on 19 October, while existing workspaces change on 2 November.
notion.com ↗Why it matters
The new billing rules, beyond earlier Notion integrations, make model selection and agent budgets an operational decision for workspace owners.
AI in the nonprofit sector2
MenoZen’s expanded 20-person study tests AI and VR acceptability
The nonprofit Menopause Society’s 20 October release highlights qualitative validation of MenoZen VR, developed by researchers at Cedars-Sinai and Mayo Clinic. Twenty women used the AI avatar Maia on a Meta Quest 3S for a 45-minute session, and all said they would use it again. The conference abstract measures usability and acceptability, not proven symptom reduction.
menopause.org ↗Why it matters
The expanded validation moves beyond an earlier eight-person pilot, while leaving clinical effectiveness for further study.
Humanity AI’s $10 million call closes Thursday morning in Sydney
Humanity AI’s previously announced Communities Leading on AI call closes at 10:59 am AEDT on 22 October, equivalent to 7:59 pm US Eastern on 21 October. Grants range from $75,000 to $1 million for up to three years, within a $10 million pool. Lead applicants must be US-based 501(c)(3) organisations, or work through an eligible fiscal sponsor, with proposals focused on US communities.
humanityai.ai ↗Why it matters
This is a final application-window reminder, not a new grant launch; the official submission deadline resolves conflicting third-party dates.