AI NewsSeptember 29, 2026

SpaceXAI opened a public beta of Team Bots, letting a company configure one AI agent per role that multiple employees can use together in Slack.
SpaceXAI launched Team Bots on September 28, now in public beta for Teams and Enterprise plans. A Team Bot draws on shared files, instructions, and connected tools like Salesforce or GitHub, while each employee's individual chats with it stay private.
SpaceXAI says its own sales bots post morning briefings and its engineering bot triages bugs and delegates fixes. One report says an internal engineering bot helped a five-person team ship over 100 pull requests a day, an unverified company claim.
Key Capabilities:

• Shared skills, private chats: The team shares a bot's knowledge, but each person's conversations stay separate.
• Slack handle: Each Team Bot gets its own handle so a whole channel can work with it.
• Approval gates: Sensitive actions require sign-off, with some controls limited to Enterprise plans.

Eleven v4 and its fast sibling, Turbo, generate speech with emotion and pacing built in, and clone a voice from just 10 seconds of audio.
ElevenLabs launched Eleven v4 and Eleven v4 Turbo on September 28. Voice cloning means making an AI speak in someone's voice after hearing a short sample; here, 10 seconds is enough. Turbo starts speaking in about 100 milliseconds (roughly a tenth of a second), a company-reported number aimed at live voice agents like phone support bots.
Eleven v4 topped Artificial Analysis's voice-model leaderboard with an Elo score of 1,319, a ranking system where a higher number means more listeners preferred it in blind comparisons. That leaderboard uses each company's own voices, so it is one measure, not the only one.
Key Capabilities:

• Wide language reach: Supports over 90 languages, including Cantonese and Mongolian.
• Voice direction: Inline tags can add laughs, whispers, or specific emotions to generated speech.
• Two speeds: Eleven v4 favors quality; Turbo trades a little quality for near-instant response.

The Manus 2.0 release adds a standalone app where each personal AI agent gets its own email, phone number, wallet, and computer to act independently.
Manus released Manus 2.0 and the Cue app on September 28. Cue agents can send messages, make payments within a set budget, and take calls, while multiple agents can share a group chat to split up a project.
Manus reports its new Cascade system uses 23.2% fewer tokens and cuts costs by 32% in internal tests, figures the company has not independently verified. Cue is free in early access with an invite code.
Key Capabilities:

• Agent identity: Each Cue agent gets its own email, phone number, wallet, and computer.
• Team delegation: Multiple agents can split a task and hand off work in a shared chat.
• New tools: Manus Studio adds a video editor and a game-development environment.

The second release in the Claude 5.5 family lands a week after Opus 5.5, aimed at everyday coding and document work at a lower cost per task.
Anthropic launched Claude Sonnet 5.5 on September 28. The company says it runs over 30% faster than Sonnet 5 and can cost up to 30% less per task, mainly because it uses fewer tokens and tool calls, not because the per-token price dropped.
Anthropic positions Sonnet 5.5 as a cheaper stand-in for Opus 5.5 on routine work like bug fixes and slide decks. It scored 70.6% on Terminal-Bench 4.0, versus 10.3% for Sonnet 5, though one independent tracker measured higher task costs at maximum effort.
Key Capabilities:

• Coding jump: Terminal-Bench 4.0 score rose from 10.3% (Sonnet 5) to 70.6%.
• First for cyber safeguards: The first Sonnet model to ship with cybersecurity protections similar to Anthropic's top-tier systems.
• Cloud reach: Live now on claude.ai and through AWS, Google Cloud, and Microsoft Azure.