SpaceXSI’s Grok Voice 3.0 tops Artificial Analysis’s speech-to-speech index

Anthropic has reworked the Claude Code desktop app for developers who keep many coding sessions going side by side. Its @ClaudeDevs account announced the change on X at 00:31 IST on 15 October. Anthropic says new sessions now open much faster, a job can be divided among several sessions, and the status of each one stays in view.

The video in the post shows the new layout. The sidebar sorts sessions under Needs input, Working and Completed, so a session waiting for an answer stands out, and sessions move between the groups as their state changes. Sessions that share out a larger task sit indented under the session they came from. A new-session screen lists the user’s projects, offers an environment picker and has one box to describe a task or ask a question.

Anthropic hasn’t published release notes or a blog post for the update, and the post gives no version number or rollout details.

Meta built an internal coding scaffold, a tool that writes and runs analysis code, on Muse Spark 1.3 and Muse Wonder, the model it introduced on 12 October. Medical researchers in fields such as cardiology and infectious disease used it for four weeks to brainstorm hypotheses, write and run all the code and draft manuscripts, keeping every scientific decision for themselves. It could reach public data and a licensed database of five million de-identified electronic health records.

The ten manuscripts range from sports medicine to oncology. One, a case series on the fungus Cladophialophora bantiana, found 21 patients with it in brain or lung samples. Each paper had feedback from peers, a biostatistician’s review and summaries from independent experts Meta contracted. Participants reported up to ten times their usual productivity in the best cases.

What the researchers found

In a report Meta hosts, the researchers call the system fast but unevenly reliable. At times it was confidently wrong with nothing to warn a reviewer, and more than once it said data was missing when it was in the dataset. It put mortality from Staphylococcus aureus bacteremia, a bloodstream infection, at 1%; an infectious disease doctor knew published studies report closer to 30%, which led to an audit of how deaths were recorded.

The studies are retrospective, using existing records only. Meta says future work on larger datasets could produce hypotheses to test in prospective trials. It hasn’t said whether researchers outside the project will get the tool.

Anthropic opened the public beta on 14 October. OpenTelemetry is a vendor-neutral standard for monitoring data: each request produces a trace, made of spans, one for every step such as a model call or a tool use. Because Claude’s spans use OpenTelemetry’s naming conventions for generative AI, tools that follow them recognise the data with no extra configuration.

There are two ways to collect traces. An admin adds a destination in the Claude Console, and the API sends spans for Messages API requests and for each Claude Managed Agents turn, model run and server tool. The Claude SDK, in Python, TypeScript, C#, Go, Java, PHP and Ruby, adds spans for tools the application runs itself.

By default, spans carry IDs, timing, status, token counts and the names of agents and tools. A Full content mode adds prompts and responses, but organisations with HIPAA, FERPA or K-12 restrictions get only metadata at Console destinations.

Anthropic lists 12 partner tools, including Datadog, Grafana Cloud, Honeycomb, New Relic and Sentry, and any tool with a standard OTLP endpoint works. API tracing runs on the Claude Platform and Claude Platform on AWS. Delivery is best effort, so Anthropic points to its Usage API for complete usage counts. Neither the announcement nor the docs mention a charge.

OpenAI notified developers on 14 October that five GPT-4 series snapshots will be removed from its API on 14 January 2027. Its deprecations page names gpt-6-luna as the replacement for all five, and fine-tuned versions of these models shut down on the same date as their base models.

• GPT-4o: gpt-4o-2024-08-06 and gpt-4o-2024-11-20
• GPT-4o mini: gpt-4o-mini-2024-07-18
• GPT-4.1: gpt-4.1-2025-04-14
• GPT-4.1 mini: gpt-4.1-mini-2025-04-14

That gives three months’ notice. The same page sets a minimum of six months for generally available models, unless safety or compliance concerns require a faster timeline, and the notice doesn’t say why this one is shorter.

An earlier batch goes first: on 23 October 2026, OpenAI removes gpt-4o-2024-05-13, gpt-4.1-nano and the older gpt-4 and gpt-4-turbo snapshots, announced on 22 April 2026. Developers can check a model’s shutdown_date field through the List models endpoint, and OpenAI says some may be able to arrange dedicated capacity through its sales team to keep using a model after it shuts down.

NVIDIA announced the commitment on 14 October. The donation is computing access worth $50 million a year for five years, set aside for research on the safety and security of AI agents, software that carries out tasks on its own.

Most of it goes to the wider UK AI safety and security research community. Part supports three groups: BOLD, the British Open-Ended Learning and Discovery Lab led by the University of Oxford; SOFAIR, the Science of Fundamental AI Research Lab led by UCL, which explores open source AI architectures; and Imperial College London, whose work includes safety engineering and robotics.

UK-based NVIDIA Cloud Partners will host the computing in the UK, but NVIDIA says provider selection, technical architecture and commercial terms are still to be agreed. The research will cover secure agents, open-model security and defences against real-world threats, including work with NVIDIA’s OpenShell runtime software and its Sentry reference design.

The same day, NVIDIA announced planned investments in five UK startups: Boltz, Foresight, Hiverge, Prima Mente and Riverlane.