AI NewsOctober 15, 2026

HeyGen Voice is the company’s first in-house voice model, built to perform a script in your own voice with your pacing and expression. It’s live on HeyGen and its API.
HeyGen Voice is built for video. It performs your script in your own voice, keeping your pacing and expression and adapting to the room acoustics of the scene. It debuted at No. 1 on Artificial Analysis’s Controlled Voice arena, where models speak using the same eight cloned voices.
API users get 50% off through Oct 31, at $15 per million characters instead of $30. Creators can keep their own voice in a video without re-recording.
Key Capabilities:

• Built for video: Performs your script in your voice, matching pacing, expression and the scene’s acoustics.
• Controlled arena: Ranks No. 1 on Artificial Analysis’s Controlled Voice arena, ahead of Qwen-Audio-3.1-TTS-Plus and Eleven v4 Turbo.
• Launch discount: 50% off the API through Oct 31, at $15 per million characters.

Max subscribers can now run Ultrafast versions of two OpenAI models inside Perplexity Computer. Pick Astra or Sol 6.1 under Custom, then set speed to Ultrafast.
Perplexity added an Ultrafast speed setting for GPT-6 Astra and GPT-6.1 Sol in Computer, its AI agent workspace for long tasks. Both generate tokens (word chunks) faster, so long Computer tasks finish sooner. It’s available only to Max subscribers.
OpenAI introduced Ultrafast on Sept 29 as a premium tier with up to 8x faster output, priced above standard. No credit multiplier or speed figure is published yet for Perplexity’s version.
Key Capabilities:

• Where to find it: Choose Astra or Sol 6.1 under Custom, then set speed to Ultrafast.
• Max only: Available to Max subscribers in Perplexity Computer.
• OpenAI’s tier: OpenAI sells Ultrafast as its fastest service tier, at higher per-token prices.

The Claude Code desktop app now starts sessions faster, makes it easier to split work across several, and shows what each is doing. A new session screen keeps recent folders one click away.
Anthropic updated the Claude Code desktop app for developers who run several sessions side by side. Starting a session is quicker, the new session screen keeps your recent folders one click away, and the sidebar shows what each session is doing.
The /btw command now opens side questions in their own tab while the main session keeps going. Claude can also spin off sub-sessions to split a big job into parts.
Key Capabilities:

• Faster starts: The new session screen keeps your recent folders one click away.
• Side questions: /btw opens questions in their own tab while the main session continues.
• Sub-sessions: Claude can spin off sub-sessions to split larger tasks into parts.

Ling-3.1-flash is a 560B-parameter model that activates about 25B per token and reads up to 1M tokens of context. It turns a rough brief into a web page, mobile UI or interactive prototype.
Turning a rough idea into a working screen takes design and coding work. Ant Group’s Ling-3.1-flash drafts pages, mobile screens and interactive 3D prototypes from a short brief, like a junior designer’s first draft.
It’s a mixture-of-experts model, so only about 25B of its roughly 560B parameters (learned settings) work on each token (word chunk). Weights are open on Hugging Face and ModelScope, so teams can run and adapt it themselves.
Key Capabilities:

• Open weights: Free to download from Hugging Face and ModelScope.
• Long memory: Supports up to 1M tokens of context, per Ant Ling.
• Work benchmarks: Ant reports 1,673 Elo on GDPVal-AA v2.1 and 75.16 on FrontierSWE.

Grok Voice 3.0 is a speech-to-speech model: audio goes in and audio comes out, with no text step in between. It’s priced at $0.08 per minute through SpaceXAI’s Voice Agent API.
SpaceX Super Intelligence launched Grok Voice 3.0, which debuted first on Artificial Analysis’s Speech-to-Speech Index with 86.9, ahead of OpenAI’s GPT-Live-1 at 83.2 and Google’s Gemini 3.8 Live at 82.6. It starts responding in under a second.
The price stays at $0.08 per minute, the same as Grok Voice 2.0. SpaceX says it already runs millions of Tesla conversations a day and resolves about 70% of Starlink support calls without a human.
Key Capabilities:

• Audio in, audio out: Answers directly in speech, without first turning the caller’s words into text.
• Strong on reasoning: Scores 98.5% on a spoken-reasoning test and 97% on a test of handling interruptions.
• Available now: Live in the Voice Agent API and Voice Agent Builder at $0.08 per minute.

Lovable’s new growth tools write the ad copy, generate images and video from your app, and build campaigns. Nothing spends money until you approve the launch.
Lovable, the AI app builder, now helps you find customers after you build. Connect a Google Ads or Facebook account in chat, and it creates ad text, images and video from your app, then builds a campaign for Google Search, YouTube, Maps, Facebook or Instagram.
You approve before anything spends, and Facebook and Instagram campaigns are created paused. Advertising is on all plans, so you can test a campaign without leaving the app.
Key Capabilities:

• Ads from your build: Writes ad copy and generates images and video using your app as context.
• Approval first: Nothing spends until you approve, and Meta campaigns start paused.
• Beyond ads: Also offers a growth plan, SEO help and Google Business Profile setup.