Weekly AI news: Pika H3 Flex, ElevenLabs SFX v3 and Gemini changes — 4 November 2026

Publisher disclosure: MUVO offers competing music, remix, and video tools. This report summarizes public sources and provider documentation; it is not an independent product test or a claim that MUVO is free of similar issues. Read our reporting method.

The week in AI: more control over creative work, more changes to track

Pika introduced H3 Flex video generation, ElevenLabs released SFX v3, and Suno and Mureka improved music-editing workflows. Runway connected its developer platform to coding agents, while Google and Anthropic set migration deadlines that developers should put on their calendars.

This roundup covers 29 October–4 November 2026, the seven Istanbul calendar dates ending today, with a research cutoff of 16:34 Europe/Istanbul (UTC+3) on 4 November. It selects the most useful product developments from MUVO’s daily reporting and rechecks the original sources. Announcement dates, available features and future commitments are kept separate.

Pika H3 Flex: editable audio is part of the video workflow

Pika launched H3 Flex through its API Club on 29 October. The model is post-trained from MiniMax H3, accepts up to 12 image, video and audio references, and generates video with synchronized audio. Its hosted workflow can return separate dialogue, score, sound-effects, Foley and ambience stems. Pika’s launch announcement.

That separation matters for editors who want to adjust the music or dialogue without treating the soundtrack as one finished mix. Native generation is 480p or 768p; 1080p, 2K and 4K finishing uses an upsampler. The released generator adapters and inference code do not include all hosted finishing features.

Availability caveat: Pika scheduled its consumer creation-product release for 2 November, but the announcement still uses future-facing language. This review confirms the API announcement, not completion of that consumer rollout. Pika’s advertised seven-second result measures transformer computation in a controlled example, not total generation turnaround.

ElevenLabs SFX v3: richer sound scenes from a prompt

ElevenLabs announced SFX v3 on 29 October, with an update on 30 October, for ElevenCreative and ElevenAPI. The company describes better handling of prompts containing several sounds, stronger musical and vocal elements, and fewer unwanted audio artifacts. These are provider claims, not MUVO benchmark results. Original SFX v3 announcement.

The release is useful for video, game and other sound-design work: an effect can describe a sequence or environment rather than a single isolated noise. ElevenLabs lists the Sound Effects tab, Flows, Studio, Music Editor and its mobile app as creation surfaces. The announcement does not establish a universal plan entitlement or a new price, and this is a sound-effects release rather than a replacement for ElevenLabs Music.

Suno and Mureka: practical improvements to existing music projects

Suno’s 29 October Studio update makes clip editing easier to find and more precise. A sidebar collects volume, pitch, loop, reverse and time-stretch controls; pitch can be adjusted in semitones and cents. New options control track visibility, the timing grid and whether imported audio follows the project tempo. Suno release notes.

There is a useful exception for musicians working with recordings: clips without a steady beat and groups of stems uploaded together retain their original timing. The update does not announce expanded Studio plan access.

Mureka’s 29 October update lets an uploaded MP4 supply audio for Remix or Audio Reference. That can remove a separate audio-extraction step when starting from a video recording. It does not turn the feature into general video editing, and the short release entry gives no plan or regional breakdown. Mureka changelog.

Analysis: both changes help with the work after the first idea—tuning, timing and bringing existing material into a project. For creators, these can matter as much as a new model name.

ChatGPT: more Work allowance and easier mobile account switching

OpenAI’s 29 October notes increase Work and Codex allowances for selected paid plans. Pro 100 and Pro 500 receive twice their previous allowance. Pro 200 users already on the higher allowance keep it; those on the lower allowance receive a doubling to the same level. The entry also announces an increase for Business Premium without specifying its multiplier. ChatGPT release notes.

On 30 October, OpenAI began rolling out account switching on iOS and Android. Up to two accounts can remain signed in; each retains separate chats, memory, settings and subscriptions. This makes personal/work switching simpler without combining their benefits. Availability can vary during rollout.

A separate future change needs attention: OpenAI’s current help page says Plus, Pro and Business will share a ChatGPT/Codex usage allowance starting in December, with token-based measurement and Instant chat excluded. MUVO first verified that guidance on 2 November; its exact original publication time is unknown. No specific December start day or numeric allowance is given. Usage-limit documentation.

Runway connects creative development to coding agents

Runway’s 2 November release introduces Dev plugins for Claude Code, Codex and Cursor. A connected agent can inspect a Runway Dev project’s models, prices, limits and credits, then use Runway’s integration skills to help build or debug media features. Runway’s original announcement.

This is most relevant to teams building video tools or automating creative pipelines. It is a developer integration, not a new video model or a grant of free generation credits. Runway’s guide distinguishes Dev MCP for development from Media MCP for generation in chat; applications making live API calls still need their own configured API credentials. Coding-agent guide.

Claude Code and Codex improve reliability; one fast mode is retiring

The week’s coding-tool releases focus on permissions, recovery and long-running work. OpenAI’s Codex CLI 0.166.0, published on 29 October, adds transcript search/selection, queue resumption and focused network diagnostics, alongside fixes that preserve approvals and permissions when resuming or forking. A CLI release does not establish an identical desktop-app update. Official Codex release.

Anthropic’s Claude Code 2.1.318, published on 4 November, exposes stored login-expiry information, warns unattended text jobs ahead of expiry, and fixes several lost-history, configuration-saving and macOS session-stall cases. Teams running unattended work should review these changes against their setup. Official Claude Code release.

A separate 2.1.316 notice says Opus 4.8 fast mode was deprecated on 30 October and will be removed on 30 November. Those fast-mode requests will error after removal rather than silently use standard mode; Anthropic directs users needing fast mode to Opus 5.5. This is a specific mode retirement, not a claim that all Opus 4.8 access ends then. Dated retirement notice.

Google: Computer Use reaches GA, while Gemini 2.5 migrations get a deadline

Google made the Computer Use tool generally available across the Gemini Interactions and GenerateContent APIs on 30 October. Developers can build workflows around screenshots and returned actions; they still supply the execution environment and appropriate permissions. This API milestone does not give every consumer Gemini account computer control. Gemini API release notes, Computer Use guide.

On 2 November, Google deprecated Gemini 2.5 Pro and Flash and removed them from AI Studio’s model selector. Its scheduled 1 December API redirects are:

Destination rates will apply. Google flags request-compatibility changes, including thinking controls and function-call validation, so developers should test actual requests rather than only replace a model name. These are future API redirects, not a description of all consumer Gemini features. Original deprecation notice.

Google Vids: an avatar launch is paused; Meet notices start rolling out

Google announced image-upload custom avatars for Vids on 2 November, then paused the rollout on 3 November. The feature is intended to turn images from a computer, Drive or Photos into reusable project avatars. At the weekly cutoff, the rollout remains TBD. Listed eligibility—including personal accounts and multiple Workspace and Google AI plans—does not establish current access to the paused feature. Announcement and pause update.

Separately, a 3 November Meet announcement describes administrator-written notices shown before joining calls, including when Gemini takes notes. The feature is off by default and began an extended rollout on 2 November for listed Workspace editions. It gives organizations a place to explain their policies; it is not a new entitlement to Gemini note-taking. Meet announcement.

Other developments worth knowing

• Mistral Large 4 weights, 29 October: Mistral released improved downloadable weights for Large 4, also called Le Chonk, following its earlier API preview. This expands hosting and customization choices. The current model page still says Public Preview, and use is governed by the model’s license; downloadable does not mean unrestricted. Weight-release announcement, model documentation.
• GPT-6 Rosalind, 30 October: OpenAI’s API changelog records a specialized biological-reasoning release for organizations approved for Rosalind Research, for internal research by authorized users. This is restricted research access, not a general ChatGPT rollout. Official API changelog.
• Gemini Notebook textbooks, 29 October: Google added hundreds of textbooks from providers including OpenStax, Open Textbook Library and S. Chand. Users can add supported titles as sources and generate study materials with source-linked answers. Google’s announcement.
• Grok image migration, effective date 2 November: An older provider notice schedules grok-imagine-image-quality to redirect to grok-imagine-image-2.0 at quality: "low". It belongs in this week because of the effective date, not a new announcement. MUVO did not run a paid generation to verify the switchover. Official migration guide.

Analysis: the useful question is what changes in your workflow

Across this selection, three practical decisions stand out. Creators gain finer control over clips, references and sound layers. Developers gain more agent integrations but also need to review permissions, model replacements and recovery behavior. Subscribers and teams should separate features they can use now from staged rollouts, paused launches and December plans.

The immediate action is to check the tools and model IDs already in use. The most consequential dates in this roundup are 30 November for Opus 4.8 fast mode and 1 December for the two Gemini API redirects. For creative production, verify actual account access before scheduling work around a newly announced feature.

Sources

The original announcements and documentation are linked beside each item. Principal sources are Pika, ElevenLabs, Suno, Mureka, Runway, OpenAI release notes, Anthropic’s versioned releases, Google’s API changelog and Mistral.

For more detail, read MUVO’s 30 October, 31 October, 2 November, 3 November and 4 November editions. Outages and recoveries remain in the incident reports.

This is a selected weekly summary, not an exhaustive launch catalog. Older announcements, undated marketing changes and rumors are excluded. MUVO publishes this reporting and offers competing creative tools. Vendor quality claims have not been independently benchmarked, and no paid generations were submitted for this article. See our reporting method.

Update timeline

• 4 November 2026, 16:34 Istanbul / 13:34 UTC: Weekly research cutoff. Rechecked original sources, grouped related developments from 29 October onward, retained the Vids pause and future migration dates, and distinguished the Grok effective date from its older announcement.