Releases, renames and access changes that affect what you can actually use. Each note links to the publisher and keeps its claims separate from our interpretation.
Selected sources checked . This is an editorial snapshot, not a live news feed or a complete release history.
Product changes
Meaningful changes to an offering, its name, interfaces or availability. Open the listing and its sources to check your exact plan and setup.
Publisher documents: Gamma's September–October release note introduces Gamma 5, connected context and changes to PowerPoint export handling.
Access and limits: These are publisher descriptions. ASE has not tested export fidelity, connector permissions or the resulting presentation quality. Credits and feature access still depend on the plan.
ASE read: Presentation comparisons should include editability and export checks, alongside generation features.
Primary documentation reviewed. No ASE hands-on test or independent benchmark is claimed.
Publisher documents: Anthropic documents general availability of the unified chat/Cowork experience for Claude Enterprise, following the September Pro/Max rollout. Agent work runs in Anthropic-managed cloud environments.
Access and limits: Enterprise administrators control access. The earlier announcement separately labels Docs, Slides and in-chat Design as paid-plan beta features; this is not proof that every feature is GA or included in Free.
ASE read: Treat Cowork as a Claude workflow and search alias, while checking individual feature entitlements.
Primary documentation reviewed. No ASE hands-on test or independent benchmark is claimed.
Publisher documents: Figma's October 6 note takes the agent out of beta. Its October 15 note separately describes code layers and Specs as open beta, code sources as closed beta, and centralized design systems as a preview.
Access and limits: Availability belongs to the individual feature. AI credits, seats and rollout conditions still matter. ASE has not tested generated code or a production deployment.
ASE read: A suite-level GA badge should not imply that all of its AI features have the same maturity.
Primary documentation reviewed. No ASE hands-on test or independent benchmark is claimed.
Publisher documents: OpenAI's October 14 notice deprecates specified GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini API snapshots, with shutdown scheduled for January 14, 2027.
Access and limits: Deprecation is a migration notice, not an already-completed shutdown. The notice names snapshots and includes corresponding fine-tunes. ChatGPT's model availability follows a separate schedule.
ASE read: Check the exact endpoint and snapshot in an integration before planning a replacement.
Primary documentation reviewed. No ASE hands-on test or independent benchmark is claimed.
Publisher documents: Letta's current documentation says the previous Letta Code Desktop app was deprecated on October 12 and no longer runs. The current Desktop offering installs the Cloud web app as a PWA.
Access and limits: The documentation does not state its publication date. The PWA needs a signed-in Cloud account; local agents use the CLI, SDK or a custom interface. Local state does not establish local model inference.
ASE read: Old native-app launch posts should not be used as a current local-desktop recommendation.
Primary documentation reviewed. No ASE hands-on test or independent benchmark is claimed.
Publisher documents: AnythingLLM's October post describes new local transcription and speaker-identification models in Desktop 1.17.0 for Mac and Windows.
Access and limits: The source gives a month, not a publication day. Its speed and accuracy statements are vendor tests, not ASE measurements. Optional cloud models and other connected features have separate data routes.
ASE read: Compare where meeting audio is processed and verify speaker labels on representative recordings.
Primary documentation reviewed. No ASE hands-on test or independent benchmark is claimed.
Publisher documents: Google announced the new name and logo, with rollout across interfaces over the following weeks. The announcement says the source-focused product remains standalone and existing shared links redirect.
Access and limits: A rename does not establish a new product, new capabilities or identical mobile features. Account, plan and source limits still need checking.
ASE read: Keep one product record and preserve NotebookLM as a search alias.
Primary documentation reviewed. No ASE hands-on test or independent benchmark is claimed.
Scientific work, robotics demonstrations and restricted previews can signal what is coming. These notes are separate from catalogue recommendations and do not admit the projects to the catalogue.
Publisher documents: Anthropic describes drug-development collaborations and early laboratory and operational results involving Claude.
Access and limits: Anthropic says the work is mostly preclinical. The published case studies do not establish clinical efficacy, regulatory approval or reproducible results for an ordinary Claude account. ASE has not independently evaluated the experiments.
ASE read: Watch scientific workflows and validation methods without presenting a medical product recommendation.
Primary documentation reviewed. No ASE hands-on test or independent benchmark is claimed.
Publisher documents: Google DeepMind announced an interactive world model with generated audio and describes an initial research preview for trusted testers.
Access and limits: Broader Ultra, API and AI Studio access is described as coming later, with a waitlist. The publisher notes drift, object-permanence and physical-interaction limitations. A demo or waitlist does not establish general access or a validated simulator.
ASE read: Track world models separately from broadly usable video tools until access and reliable workflows are established.
Primary documentation reviewed. No ASE hands-on test or independent benchmark is claimed.
Publisher documents: Google's developer article demonstrates robot control, real-to-simulation reconstruction, video labeling and iterative hardware experiments using Gemini 4 Argon.
Access and limits: These are publisher demonstrations with particular hardware, calibration and software. ASE has not reproduced them or verified safe deployment. The article does not establish a universal robot controller or access to a complete robotics product.
ASE read: Physical AI deserves monitoring, with hardware and safety validation kept distinct from model demonstrations.
Primary documentation reviewed. No ASE hands-on test or independent benchmark is claimed.
Publisher documents: Google DeepMind announced a portal, API and Antigravity skill for exploring predicted molecular effects of single-letter human DNA variants.
Access and limits: The announcement describes academic research access and explicitly says AlphaGenome is not clinically validated or approved. Prediction scores are not diagnoses. ASE has not assessed scientific accuracy or account-specific access terms.
ASE read: Follow specialized scientific tools without treating research availability as clinical readiness or a general assistant recommendation.
Primary documentation reviewed. No ASE hands-on test or independent benchmark is claimed.