AI NewsSeptember 24, 2026

Alibaba launched Qwen Intelligence, a full-stack agentic AI platform for phone makers with three mobile agents for planning, app control, and image creation, debuting first on HONOR's upcoming Magic9 series.
Alibaba launched Qwen Intelligence on September 22, a full-stack agentic AI platform for smartphone makers rather than a phone of its own. It opens with three agents: a Planner for breaking down tasks, a Mobile-Use agent that operates apps, and a Creative agent for images.
HONOR is the first partner, integrating it into MagicOS ahead of the Magic9 series launching September 28. Alibaba reports a 90% end-to-end completion rate on complex tasks and says its Creative Agent cuts image generation from 100 steps to just 8.
Key Capabilities:

• Full agent stack: Planning, app operation, and image creation bundled for phone makers to build on.
• Fast image generation: Creative Agent cuts image steps from 100 to 8, producing a result in about 3 seconds.
• HONOR first: Debuts on HONOR's Magic9 series and Robot Phone, launching September 28.

Anthropic moved Claude Code's cloud sessions out of research preview to general availability, letting the AI keep coding on Anthropic's infrastructure after you disconnect, with free credits for Pro and Max subscribers through October 7.
Anthropic moved Claude Code cloud sessions out of research preview on September 23, letting Claude keep working on bug fixes even after you close your laptop. Pro subscribers get a one-time $100 credit, Max subscribers get $250, claimable through October 7.
Sessions start from the web, mobile app, desktop app, or claude --cloud in the CLI, and require a GitHub connection since each runs on its own branch and repo copy. The credit is separate from regular usage limits and applies automatically once a session begins.
Key Capabilities:

• Runs without you: Coding sessions continue on Anthropic's infrastructure after you disconnect.
• Free credits: $100 for Pro, $250 for Max, claimable through October 7.
• GitHub required: Each session runs on its own branch and repository copy.

Google added local model support to the Antigravity SDK, letting developers run Gemma 4 26B directly on their own hardware for zero API costs, total data privacy, and no internet connection required.
Google updated the Antigravity SDK on September 23 to support local models, with initial support for Gemma 4 26B running through Google AI Edge's LiteRT. Developers can now execute agentic coding workflows entirely offline on their own GPU, with zero API costs.
A hybrid demo shows Gemini 3.8 Flash acting as cloud planner while local Gemma 4 instances do the heavy lifting, patching vulnerable code without any source ever leaving the machine. Over 97% of tokens in that workflow were processed locally.
Key Capabilities:

• Fully offline: Runs Gemma 4 26B on a local GPU with no internet required after setup.
• Zero API cost: Local workflows skip cloud inference charges and rate limits entirely.
• Hybrid option: A cloud model can plan while local Gemma instances execute, keeping code private.

OpenAI upgraded ChatGPT Voice with plugin support for email, calendar, and Slack, running on GPT-6 Astra, Sol, and Luna, and now usable inside ChatGPT Work to create documents and spreadsheets just by talking.
OpenAI upgraded ChatGPT Voice on September 23, adding plugin support for tools like email, calendar, and Slack, so a spoken request can pull from connected apps. Voice now runs on GPT-6 Astra, Sol, and Luna, the same models powering ChatGPT's text and Work modes.
It's also now usable inside ChatGPT Work on web and mobile, letting people create documents, decks, sites, and spreadsheets or handle complex browser tasks just by talking. The rollout is global, though some users report bugs with Voice becoming unresponsive in Codex.
Key Capabilities:

• Plugin-connected: Talk to email, calendar, and Slack directly through voice, no typing needed.
• Upgraded models: Runs on GPT-6 Astra, Sol, and Luna instead of older voice-specific models.
• Full Work integration: Build docs, decks, sites, and spreadsheets in ChatGPT Work by speaking.

Fish Audio launched Drama 3 preview, a highly controllable text-to-speech model that follows plain-language direction on tone and pacing, supports mid-sentence voice shifts, and can regenerate a single word.
Fish Audio launched Drama 3 preview on September 23, pitched as its most controllable TTS model yet. Instead of learning audio tags or SSML markup, you describe the tone, pacing, and character in plain language, and the model follows the direction.
It supports shifting voices mid-sentence, generating a multi-character scene in one pass, and regenerating a single word without redoing the whole take. It joins Fish Audio's S2 family as one of only two models supporting multi-speaker dialogue; access is gated behind a request form.
Key Capabilities:

• Plain-language direction: Describe tone and pacing in words instead of learning audio tag syntax.
• Precise repairs: Regenerate a single word inside a take without redoing the whole recording.
• Multi-character scenes: Generates full dialogue scenes with distinct voices in one pass.

Google launched Gemini 3.8 Flash TTS and Flash-Lite TTS, letting developers create custom voices from a text prompt or clone one from a 30-second sample, with line-by-line control over pacing and emotion.
Google launched Gemini 3.8 Flash TTS and Flash-Lite TTS on September 23, its most expressive audio models yet. Developers can design new voices from a natural-language prompt, pick from over 2,000 ready-made options, or clone a voice from a 30-second sample.
Flash TTS targets high-fidelity creative work like gaming and audiobooks with full line-by-line direction; Flash-Lite is built for cost-efficient scale, adjusting tone automatically for real-time voice agents. Both include built-in watermarking, and Flash TTS now tops Hume AI's voice benchmark.
Key Capabilities:

• Voice design from text: Describe a voice in plain language and get a custom character voice.
• Line-by-line direction: Control pacing, emotion, and dialect shifts for each line individually.
• Built-in safety: Watermarking included, plus a mandatory consent check for voice cloning.

Alibaba released Qwen-Audio-3.1, a five-model audio stack covering speech recognition, text-to-speech, and real-time conversation, with API price cuts up to 95% for recognition, 85% for realtime, and 70% for TTS.
Alibaba released Qwen-Audio-3.1 on September 23, a five-model lineup covering ASR, TTS, and realtime conversation, plus two additions: TTS-Next for audio creation and ASR-Next for deeper understanding. ASR-Next adds multi-speaker diarization, timestamps, emotion detection, and ambient sound recognition.
Alongside the launch, Alibaba slashed API prices across the lineup: speech recognition drops up to 95%, realtime conversation about 85%, and text-to-speech roughly 70%. TTS-Next pairs a language model with diffusion to generate voice, sound effects, and background audio in a single pass.
Key Capabilities:

• Massive price cuts: Speech recognition down up to 95%, realtime conversation about 85%.
• Multi-speaker aware: ASR-Next identifies speakers with timestamps and detects emotions and ambient sound.
• One-pass creation: TTS-Next generates voice, effects, and background audio together for audiobooks and podcasts.