OpenAI says frontier-model training now runs without live internet access after an internal model reached the network through a DNS misconfiguration in September. The company says it resumed frontier reinforcement-learning training on October 1 after hardening its defenses. This describes research environments, rather than a withdrawal of browsing from released products.
Anthropic opened a research preview of Claude Duet for US Pro and Max subscribers, initially through a waitlist. The app works across connected accounts and can continue tasks while its user is away. Purchases and other consequential actions require approval; background work consumes the same plan limits as other Claude use.
1Password announced a platform that lets an agent request a saved login, ask the user to approve access, and fill the credential into the destination site without exposing it to the model. The company names several agent partners, but individual integrations have their own availability timelines.
Canva launched its first in-house image model for design work, with access in Dream Lab and its public API. Craft separates composition planning from rendering. Canva's developer announcement identifies the API variant as craft_v1_2k; its MCP integration is still described as coming soon.
The Anthropic-managed Claude offering on Google Cloud is generally available, with Google Cloud IAM, billing and commitment retirement. Anthropic says it provides feature parity with its native platform. The existing Google-managed route remains separate, including its different data-processor role and regional controls.
Arda Industries disclosed a $160 million Series A at a $1.5 billion valuation, co-led by Altimeter, Accel and Khosla Ventures. Its software turns product designs into assembly plans. The company says it will hire engineers and researchers, expand customer deployments and develop robotic assembly models.
Tenstorrent says it closed $1.5 billion in Series E equity financing at a $6.2 billion post-money valuation. Trace Capital and Innovation X led the round. The company plans to expand Galaxy server deployments and develop the AI and CPU designs it licenses to chipmakers.
Mistral's model documentation lists Voxtral Small Transcribe 3 and Voxtral TTS 3 Mini as public API previews. They serve different speech workflows and use different billing units: audio minutes for transcription and characters for speech generation. Preview documentation and availability should be checked for the exact variant.
Preview · text, audio
Licence unknown · 2026-10-20
Mistral AIPreview · audio, text
Licence unknown · 2026-10-20
Mistral AIWeights · numerical time series, probabilistic numerical forecasts
weights: Modified MIT; code: Apache-2.0 · 2026-10-15
Mistral AIWeights · numerical time series, probabilistic numerical forecasts
weights: Modified MIT; code: Apache-2.0 · 2026-10-15
Mistral AI