Daily AI news, sourcedOfficial news, first rumor reports and substantive progress. Each entry keeps the information published at that time.
- Perplexity rolls out new voice mode for Pro and Max users
Perplexity CEO Arav Srinivas announced a new voice mode rolling out for all Pro and Max users, per his official account.
- Perplexity introduces JOLT, a post-training method that matched GRPO on MATH with 6.5x fewer training tokens
Perplexity says JOLT trains one model as both teacher and student, with the teacher learning to give feedback the student can use. On MATH it matched GRPO accuracy using 6.5 times fewer training tokens.
- Perplexity Agent API adds GPT-6 Astra and DeepSeek V4 Pro, and older Claude, Gemini and Grok models are no longer listed
Perplexity's Agent API models page now lists openai/gpt-6-astra at $10 per million input tokens and $50 per million output tokens for inputs up to 272k tokens, with flex and priority service tiers, and perplexity/deepseek-v4-pro-0813 at $1.32 per million input tokens, $3.96 output and $0.044 for cached reads. Older models including Claude Opus 4.8 and Grok 4.3 no longer appear in the model list.
- Perplexity open-sources SPACE, the Firecracker microVM sandbox that runs every Perplexity Computer task
Perplexity has released SPACE (Sandboxed Platform for Agentic Code Execution) on GitHub under the Apache 2.0 license. Written in Rust, it gives each agent task its own Firecracker microVM, keeps credentials out of the guest by injecting them at the network layer, and enforces egress policy through a per-sandbox gateway; it needs a Linux x86_64 host with KVM. Perplexity says median sandbox creation fell from 185 ms to 60 ms versus its previous provider. In its own red-team study, none of 108 runs across 9 models escaped to the host, while 11 of 54 partial-network runs bypassed the egress policy before fixes and none after.
- Perplexity details how its SPACE sandbox uses cryptographic identity to route requests across millions of daily executions
Perplexity says SPACE, its sandbox for running code, handles millions of executions a day with an error rate below 0.0002%. A new engineering post explains how it uses cryptographic identity to keep each request tied to the right execution as sandboxes pause, resume and move.
- Perplexity explains the Git-forge filesystem behind Perplexity Projects for agents
Perplexity published a post on how it built the filesystem behind Perplexity Projects as a Git forge for agents. Each agent works in its own sandbox and commits to a shared repository, so people and other agents can build on its work across sessions.
- Perplexity rolls out inline charts, diagrams and visualizations in Computer
Perplexity CEO Aravind Srinivas announced the rollout of inline charts, diagrams and visualizations in Computer, including context-aware financial charts.
- Perplexity Agent API adds Profiles, Skills and managed connectors
Perplexity CEO Aravind Srinivas announced that the Agent API now supports Profiles, Skills and managed connectors.