AI Product Updates: New Features and Releases

A new Luna model is live on its cloud platform, with $15 in credits for sign-ups, and the author asks users to report what doesn't work.

A multi-user, multi-agent collaboration platform offers real-time shared runtime and cloud saving, with an invite code for trying it out.

DeepSeek introduced an agent training system that can run up to 380,000 sandboxes at once and is designed to contain "agent misbehavior."

Superwhisper, previously focused on the Apple ecosystem, has released an Android version of its voice transcription tool.

Listen Labs launched pricing tests with Gabor Granger as its first method, where each point on the demand and revenue curves comes with an AI-moderated interview explaining the why, now live for all users.

p0 and Render released an open-source research agent that splits a question into smaller tasks, uses Parallel Search and Extract with independent retries on Render Workflows, and writes a single report with sources.

Claude Code's next version, 2.1.281, is about to be released, with no details given on what changed.

Fish Audio has launched a preview of Drama 3, which takes plain-language descriptions of tone, pacing and character and can switch voice mid-sentence, generate multi-character scenes, or fix a single word.

Genspark has added GenCode to its Super App, letting users run Claude, GPT, DeepSeek or open-weight models from one account without API keys and pick a model per task.

OnSoloAI has launched Solo Video, which generates 15-second videos in one pass and claims fewer failed generations, with members getting 3-10 days of unlimited use depending on tier and 150 spots per tier.

ZeroDrift has released Anchor, a scanner that flags risky lines in a wealth advisor agent's replies, cites the exact rules, rewrites the response and blocks it if it still fails.

A reposted claim says three new plans are coming to the OpenAI Platform so users can build and host apps, but the post gives no details on features or limits.