AI news

AI news, with the practical details

Dated updates from official sources. Understand what was released, where it is available and what to check before changing your workflow.

Announced

Copilot retires older models: check the replacement before your next task

GitHub’s October 19 notice retires selected older models across most Copilot experiences and suggests replacements. Check the model selector and organization policy before starting work. The general documentation still lists Grok 4.5, conflicting with the dated retirement notice.

Coverage published 20 Oct 2026 · Sources checked 20 Oct 2026
Announced

Mercury 3 promises shorter waits, with a launch discount and a migration date

Inception launched Mercury 3 on October 19 with a 1M-token context window. Its own platform lists launch rates of $0.10 input and $0.35 output per million tokens. The speed figure is a vendor measurement. Existing Mercury 2/2.5 users should check the November 15 access deadline and enterprise exceptions.

Coverage published 20 Oct 2026 · Sources checked 20 Oct 2026
Announced

MiniMax M3.1 Flash cuts token prices, but still needs a careful migration

MiniMax’s October 19 M3.1 Flash rollout lists $0.15 input and $0.60 output per million tokens, with $0.015 cache reads. API, M Plan and Code access are distinct routes; weights are promised later. Test thinking handling and actual task cost before migrating.

Coverage published 20 Oct 2026 · Sources checked 20 Oct 2026
Announced

PixelDojo turns one product photo into a starting point for holiday ads

PixelDojo’s October 19 update adds seasonal effects and Sensei recipes, including ad sets from a product photo and a selfie-plus-product unboxing format. Review the recipe, quoted cost and product accuracy before generating or delivering an ad.

Coverage published 20 Oct 2026 · Sources checked 20 Oct 2026
Announced

MorningAI brings Outlook and client files into one agent task

MorningAI now connects its Agent to Microsoft 365. Start with one client brief and a draft reply, then inspect the source, recipients and quote before approving further work.

Coverage published 19 Oct 2026 · Sources checked 19 Oct 2026
Announced

PixelDojo lets an AI assistant find an effect before paying to run it

PixelDojo’s October 18 update brings named photo and video effects to Sensei, its MCP server and API. pixeldojo_effects lists required inputs and prices without generation credits; pixeldojo_effect runs the chosen effect at its listed engine rate, charged on success. Confirm the quote before a generation.

Coverage published 19 Oct 2026 · Sources checked 19 Oct 2026
Announced

DeepSeek refreshes Flash and opens separately billed web tools

DeepSeek’s October 17 release updates V4.1 Flash across web, app and API, keeping the deepseek-flash name and model prices. Search and Fetch are separate APIs, each listed at $0.75 per 1,000 successful requests. Selected server-side tool calls have a temporary fee waiver; model tokens remain separately billed.

Coverage published 18 Oct 2026 · Sources checked 18 Oct 2026
Announced

ChatGPT Slides keeps the deck editable after the first draft

OpenAI’s October 16 Slides update lets eligible Pro, Business and Enterprise users with Space access create and edit presentations together. Web exports include PowerPoint, PDF and Google Slides. Mobile supports conversational creation and viewing; direct element editing uses web or desktop.

Coverage published 17 Oct 2026 · Sources checked 17 Oct 2026
Announced

Cursor cloud agents can leave a canvas your team can inspect

Cursor’s October 16 update lets cloud agents create interactive canvases. Published shares are read-only snapshots for teammates, not public editable websites. Sharing needs an eligible paid plan, team membership and a privacy mode that permits storage. Cloud-agent model usage is billed separately.

Coverage published 17 Oct 2026 · Sources checked 17 Oct 2026
Announced

OpenAI lets you describe a voice, then reuse it in GPT-Live

OpenAI made prompt-to-voice available to all API customers on October 16. Describe an adult voice, save its ID and reuse it in GPT-Live. Creation is free, but listening and speech use regular session billing. These voices do not work with the separate Text to Speech, Realtime or Chat Completions APIs.

Coverage published 17 Oct 2026 · Sources checked 17 Oct 2026
Announced

FLUX 3 Image Klein brings ten-image references and open weights

Black Forest Labs released FLUX 3 Image Klein on October 15: a 5.6B image generator for text-to-image, editing and up to ten reference images, with open weights and an API. Local setup needs more than the checkpoint. Bounding boxes guide placement rather than hard-mask edits, and commercial use depends on the licence terms.

Coverage published 18 Oct 2026 · Sources checked 18 Oct 2026
Announced

Opus 5.5 fast mode drops its access request, keeps its premium price

Anthropic’s October 15 update removes the research-preview access request and beta-header requirement for Opus 5.5 fast mode. It uses the same model at premium rates: $8 input and $40 output per million tokens. The speed claim concerns output generation, not a guaranteed faster completed task.

Coverage published 17 Oct 2026 · Sources checked 17 Oct 2026
Announced

Suno adds shared workspaces, while model sharing needs Premier

Suno’s October 15 web release adds shared workspaces for Pro and Premier, with access for invited collaborators. Sharing custom models requires Premier. Older help text still says models cannot be shared, so use the dated release and verify the controls available in your account.

Coverage published 17 Oct 2026 · Sources checked 17 Oct 2026
Announced

Adobe separates faster local editing from faster cloud media search

Adobe’s October 15 announcement describes RTX Spark optimization for creative apps and GPU processing for Frame.io. NVIDIA documents phased rollout, with Photoshop and Premiere AI optimizations in beta and Substance updates coming later. Local editing and cloud media search are different improvements.

Coverage published 16 Oct 2026 · Sources checked 16 Oct 2026
Announced

ChatGPT adds new Live voices, with a screen-sharing tradeoff

OpenAI’s October 15 update rolls out GPT-Live-1.1 and its smaller Quip model in ChatGPT Voice, with new voices and accents. Availability varies. Live has plan-specific limits and does not support video or screen sharing; Advanced retains those supported mobile features.

Coverage published 16 Oct 2026 · Sources checked 16 Oct 2026
Announced

GitHub expands AI security checks, but a finding will not block a merge

GitHub made AI Scan and agentic autofix generally available on October 15. AI Scan gives advisory pull-request findings that cannot block merges. Agentic autofix handles eligible default-branch alerts and proposes draft pull requests. Their access requirements differ, and both consume Code Security AI Credits.

Coverage published 16 Oct 2026 · Sources checked 16 Oct 2026
Announced

GPT-Live can ask Decisions to act while you are still speaking

The October 15 API update lets GPT-Live call Decisions directly or trigger it from transcript changes. The application still carries out and verifies actions. Decisions pricing is separate from the cost of the Live conversation; this is not the ChatGPT Voice model rollout.

Coverage published 16 Oct 2026 · Sources checked 16 Oct 2026
Announced

YouTube Music puts AI labels on the player, with more detail upstream

YouTube began rolling out player-page labels for music created primarily with generative AI on October 14. Music partners can deliver fully, partly or no-gen-AI declarations through supported metadata routes. Those categories do not establish that every partially generated track gets the same public badge. Keep a record of how each release was made.

Coverage published 18 Oct 2026 · Sources checked 18 Oct 2026
Announced

Copilot adds HydraFusion Max for more demanding coding tasks

GitHub’s October 14 update adds HydraFusion Max and changes how models draft and review coding work. Both options remain research previews. Extra model passes can consume more AI credits; compare one bounded bug fix and inspect workspace edits before adopting it.

Coverage published 15 Oct 2026 · Sources checked 15 Oct 2026
Announced

Google and Singapore design scam detection to stay on the phone

Google DeepMind and Singapore’s GovTech describe a two-stage scam detector using local Gemma analysis on the phone. The October 14 announcement plans a phased app rollout with user consent. It does not establish that a public app or Malaysia access is already available.

Coverage published 15 Oct 2026 · Sources checked 15 Oct 2026
Announced

Moondream 3.1 reaches Prime Inference after acquisition

Prime Intellect’s October 13 announcement makes Moondream 3.1 available through Prime Inference. Credits for Moondream Cloud customers were promised for the following days. Verify the endpoint, credit balance and platform pricing before migrating a visual workflow.

Coverage published 15 Oct 2026 · Sources checked 15 Oct 2026
Announced

Cursor opens Origin’s app marketplace to more workflows

Cursor’s October 13 Origin marketplace brings deployment, CI, review and other integrations to Origin repositories. Public app submissions are open, but marketplace listing requires approval. Start with one useful workflow and review the app’s repository scope.

Coverage published 14 Oct 2026 · Sources checked 14 Oct 2026
Announced

Gemini 4 Argon goes GA in the API and Copilot

Gemini 4 Argon went GA on October 13 in Google’s API and GitHub Copilot. API users need to review caching and thinking settings; Copilot access depends on plan, rollout and policy. Standard API introductory prices end after November 30.

Coverage published 14 Oct 2026 · Sources checked 14 Oct 2026
Announced

ElevenLabs adds distance and reverb to generated speech

ElevenLabs’ October 12 update adds optional audio effects to speech and dialogue API requests, including distance, reverb and panning. Effects do not change request cost. Stereo-producing effects reject certain output formats; compare a small cue in your mix before processing a whole script.

Coverage published 15 Oct 2026 · Sources checked 15 Oct 2026
Announced

Descript lets guests record without booking a call

Descript’s October 12 roundup adds asynchronous recording requests on Enterprise drives, with a script, expiry and audience options. It also adds drive-wide search and Underlord file lookup. Check drive access, then prepare one short contribution brief before creating a request.

Coverage published 14 Oct 2026 · Sources checked 14 Oct 2026
Announced

HelixBench checks whether AI sound follows the camera

Noiz AI’s repository dates the public HelixBench release to October 12. It evaluates generated audio/video across six separate metrics, including timing and spatial direction. The HelixWorld preview is older. Compare matching metrics and subsets rather than treating one score as overall quality.

Coverage published 14 Oct 2026 · Sources checked 14 Oct 2026
Announced

Muse Wonder arrives for builders before the wider Muse rollout

Meta announced Muse Wonder on October 12 for the Meta Model API and Muse Code, with Wonder Full for Muse Code subscribers. The wider Muse rollout is still planned, first for subscribers. Check which model your current surface offers before trying a bounded task.

Coverage published 14 Oct 2026 · Sources checked 14 Oct 2026
Announced

BaoCut adds local music for non-commercial video projects

BaoCut’s October 12 release adds local music and video soundtracks. Its release notes restrict the bundled music model’s output to non-commercial use. Try it on a personal project; use music with suitable rights for paid work.

Coverage published 13 Oct 2026 · Sources checked 13 Oct 2026
Announced

Copilot for JetBrains adds tool controls and raises its IDE minimum

GitHub’s October 10 Copilot for JetBrains update adds a setting to stop MCP servers starting automatically and a diagnostic Fix action. Enterprise admins can set a default agent model. The plugin now requires JetBrains IDE 2025.2 or later. Check compatibility, then review one proposed repair.

Coverage published 11 Oct 2026 · Sources checked 11 Oct 2026
Announced

Anthropic cuts live internet access in internal AI tests

Anthropic’s October 9 report describes unintended actions and an internal-evaluation internet pause. It does not announce a customer browsing shutdown. When using a connected agent, define what happens if a required source or tool fails.

Coverage published 12 Oct 2026 · Sources checked 12 Oct 2026
Announced

Cursor Origin changes how integrations recognise automated reviews

Cursor’s October 9 Origin API change identifies automated actions by their service accounts instead of the shared origin-cursor-managed-actor identity. Integration owners should match serviceAccount.id or type and use returned account IDs in author filters. Origin remains Early Beta.

Coverage published 11 Oct 2026 · Sources checked 11 Oct 2026
Announced

Codex suggests your next message, and leaves the send button to you

Codex’s October 9 composer predictions beta can suggest a complete next message after a response. Adults on personal Pro can use it in supported desktop local or SSH threads. Tab accepts text; review it before sending. Beta prediction generation uses no credits, but sent messages follow normal billing.

Coverage published 10 Oct 2026 · Sources checked 10 Oct 2026
Announced

ElevenCreative lets you bring the same brand brief to every image

ElevenCreative’s October 9 Brand Kit stores visual assets and rules for @ reuse across Image & Video, Flows and its agent. Confirm kit access in your workspace, check the generation estimate and try one image before extending a client brief to video. Voice and sound support remains planned.

Coverage published 10 Oct 2026 · Sources checked 10 Oct 2026
Announced

Mistral can host your workflow worker, with a preview limit

Mistral’s October 9 Managed Deployments public preview builds and runs workflow workers from GitHub. It requires a paid plan: Pay-as-you-go allows three concurrent deployments and Enterprise ten. Check the workspace’s costs and limits, then validate one small workflow before relying on hosted execution.

Coverage published 10 Oct 2026 · Sources checked 10 Oct 2026
Announced

Google’s new Gemini agent comes with an early-access catch

Google announced its Gemini agent on October 8 for connected business work, including persistent cloud tasks and model choice. Its smaller-business post specifies early access, with broader availability undated. Confirm organizational access and costs before trying a source-linked draft that a person reviews before use.

Coverage published 10 Oct 2026 · Sources checked 10 Oct 2026
Announced

GPT-6.1 Sol Ultrafast puts a price on shorter waits

GPT-6.1 Sol Ultrafast is a faster Responses API service tier added October 8, using the same model. Its short-context price is $12 per million uncached input tokens and $60 per million output tokens. Check total task time and cost before switching.

Coverage published 9 Oct 2026 · Sources checked 9 Oct 2026
Announced

Suno Studio gives awkward audio clips a better fit

Suno’s October 8 browser Studio update adds time stretching and shaped fades, with improved downbeat detection. Studio is included in Premier. Try a duplicate of one short section and listen for unwanted changes before stretching a whole arrangement.

Coverage published 9 Oct 2026 · Sources checked 9 Oct 2026
Announced

Copilot can fence off local commands, if you enable it

Copilot local sandboxing became generally available October 7, with no extra local-sandbox charge. It is off by default. Enable it in the surface you use, inspect allowed files and network access, and check host requirements before letting commands run.

Coverage published 9 Oct 2026 · Sources checked 9 Oct 2026
Announced

ChatGPT answers can now become interactive tools

Intelligent UI lets ChatGPT combine text, visuals and interactive tools in its answers. The October 7 rollout starts with listed paid plans; Free and Go expansion starts October 8. Check Chat access and verify the facts behind any generated interface.

Coverage published 8 Oct 2026 · Sources checked 8 Oct 2026
Announced

Haiku 5.5 makes small AI jobs cheaper, with a migration catch

Haiku 5.5 launched October 7 with lower API rates for frequent, bounded tasks. Short-prompt input costs $0.10 and output $0.50 per million tokens; longer prompts use a higher band. Recount tokens and migrate request settings before switching.

Coverage published 8 Oct 2026 · Sources checked 8 Oct 2026
Announced

ChatGPT can now turn uploaded recordings into working notes

ChatGPT’s October 6 audio-upload update lets paid users work with existing recordings in a conversation. For an interview, start with draft notes and verify quotations against the audio before publishing.

Coverage published 7 Oct 2026 · Sources checked 7 Oct 2026
Announced

Cursor puts your local coding agent within reach of your iPhone

Cursor’s October 6 update lets you see and reply to local coding agents from its iOS app. Pair with the desktop app first and keep the original computer awake and connected; this does not move the task into the cloud.

Coverage published 7 Oct 2026 · Sources checked 7 Oct 2026
Announced

GitHub’s ReviewBench asks whether AI code review finds useful problems

GitHub launched ReviewBench’s research preview on October 5 to assess AI code-review findings against a public pull-request corpus. Its useful question is what a reviewer catches and how much noise it creates, rather than how many comments it writes.

Coverage published 7 Oct 2026 · Sources checked 7 Oct 2026
Announced

Copilot CLI adds settings commands and keeps your latest prompt

Copilot CLI 1.0.92, released October 5, adds commands for inspecting and changing settings plus a Ctrl+E local/cloud picker before a conversation. GitHub also reports fixes that preserve the latest prompt during context compaction. Review your settings and environment before resuming a task.

Coverage published 6 Oct 2026 · Sources checked 6 Oct 2026
Announced

Suno Albums turns a collection of songs into one release

Suno’s October 5 Albums update lets you present your own songs as one release on web, iOS and Android. Publishing notifies followers and also makes included draft songs public. Share the draft for feedback and check every track before releasing it.

Coverage published 6 Oct 2026 · Sources checked 6 Oct 2026
Announced

GitHub Copilot adds review API controls and a Balanced default

GitHub’s October 2 announcement adds REST and GraphQL requests for Copilot code reviews, including an effort choice. Default effort became Balanced on September 28; explicit Lite settings remain. Balanced uses more AI credits, so inspect your settings before automating repeated reviews.

Coverage published 5 Oct 2026 · Sources checked 5 Oct 2026
Announced

Runway adds Seedance drafts and Ideogram 4.5 in MCP

Runway’s October 2 update adds 480p Seedance 2.5 drafts in Agent and Ideogram 4.5 image generation through MCP. Use drafts to review shot direction before enhancement; check your plan and the complete generation cost first.

Coverage published 4 Oct 2026 · Sources checked 4 Oct 2026
Announced

ChatGPT adds multi-page PDF scanning on iOS

ChatGPT’s October 1 iOS rollout adds consecutive page capture combined into one PDF. Open the in-app camera, use its three-dot menu and select Scan. Inspect the captured pages and verify important details before relying on a summary.

Coverage published 5 Oct 2026 · Sources checked 5 Oct 2026
Announced

Suno Speech beta brings spoken audio and music into one track

Suno’s October 1 Speech beta creates spoken audio with original background music from text and style direction. Release notes list web and mobile availability. Try a short original script and listen for accent, pacing and intelligibility before planning a larger project.

Coverage published 5 Oct 2026 · Sources checked 5 Oct 2026
Announced

Firefly AI Assistant expands access and starts using credits

Adobe documents broader Creative Cloud access to Firefly AI Assistant from October 1, alongside credit use for certain paid-plan actions. The assistant remains a web beta with account exceptions. Check access and action costs before handing it a large creative job.

Coverage published 4 Oct 2026 · Sources checked 4 Oct 2026
Announced

Gemini Skills lets your writing instructions travel between chats

Google announced Gemini Skills on September 30: reusable instructions you can combine in an existing chat. Current help describes a gradual personal-account rollout for adults with Keep Activity on, with chat skills available without an AI subscription. Start with one writing task and keep Gems that still need unsupported tools.

Coverage published 6 Oct 2026 · Sources checked 6 Oct 2026
Announced

Gemini 4 Argon benchmarks: where it leads, where it falls behind

Gemini 4 Argon is a strong candidate for complex document and agent work. It leads the checked Vals Index, but other tests favor different models. Choose a benchmark that resembles your job, then compare usable results, total cost and review time.

Coverage published 2 Oct 2026 · Sources checked 2 Oct 2026
Announced

GPT-6.1 Sol reaches the API, Work and Codex

OpenAI released GPT-6.1 Sol in the API on September 29 and began its Work and Codex rollout. Standard short-context API rates are $2 per million input tokens and $10 per million output tokens. Use Responses for tools and compare completed work before replacing Astra.

Coverage published 4 Oct 2026 · Sources checked 4 Oct 2026
Announced

Eleven v4 and Turbo split narration from live voice work

Eleven v4 targets expressive content such as narration; v4 Turbo targets interactive voice. Both are available across ElevenCreative, ElevenAgents and ElevenAPI. Check the correct API path, changed voice controls and cross-language accent behavior before migrating.

Coverage published 4 Oct 2026 · Sources checked 4 Oct 2026
Announced

Claude Sonnet 5.5: what the September release means

Anthropic announced Sonnet 5.5 on September 28. Understand the model name, stated improvements and the checks to make before changing your coding workflow.

Coverage published 29 Sept 2026 · Sources checked 29 Sept 2026
Announced

Claude Opus 5.5: release details and practical next steps

Opus 5.5 was announced on September 22. Separate the confirmed release from vendor performance claims, and plan a useful comparison before switching tools.

Coverage published 29 Sept 2026 · Sources checked 29 Sept 2026
Announced

GPT-Live API: voice conversations with a separate backend

GPT-Live 1 became generally available in the API on September 10, 2026. It separates spoken interaction from backend reasoning. The $0.05-per-minute voice-session charge excludes backend model, tool and infrastructure costs.

Coverage published 29 Sept 2026 · Sources checked 29 Sept 2026
Announced

GPT-6 Astra: model, API and workflow changes explained

GPT-6 Astra launched in the API on September 3, 2026. It accepts text and images and produces text. Tool calling requires the Responses API, so an existing tool integration may need more than a model-name change.

Coverage published 29 Sept 2026 · Sources checked 29 Sept 2026