- GPT-Realtime-2: speech-in/speech-out model with GPT-5-class reasoning, parallel tool calls, spoken preambles like 'let me check that,' and a 128K context (up from 32K).
- GPT-Realtime-Translate: live speech-to-speech translation across 70+ input languages and 13 output languages, designed to keep pace with the speaker.
- GPT-Realtime-Whisper: streaming speech-to-text that emits text as the user is still talking — aimed at captions, meeting notes, and live agents.
- GPT-Realtime-2 scores 15.2pp higher than GPT-Realtime-1.5 on Big Bench Audio; Zillow reports a 26pp jump in call success rates in early testing.
- Pricing: $32/$64 per 1M audio input/output tokens for GPT-Realtime-2; $0.034/min for Translate; $0.017/min for Whisper.
- One-year impact paper: the Gemini-powered coding agent has been deployed on production scientific and engineering workloads at Google and partners.
- Genomics: cut variant detection errors in PacBio's DeepConsensus pipeline by 30%.
- Energy: lifted feasible-solution rates for an AC Optimal Power Flow GNN from 14% to 88%.
- Quantum: produced circuits with 10x lower error than hand-tuned baselines for molecular simulation on Google's Willow chip.
- Earth AI natural-disaster risk model gained 5pp average accuracy across 20 categories (wildfires, floods, tornadoes, etc.).
- Codename Hatch — an autonomous agent for Facebook, Instagram, WhatsApp, and Threads that summarizes chats, tracks competitors, and posts on trending topics.
- Currently powered by Claude Opus 4.6 and Sonnet 4.6; Meta plans to swap in its own Muse Spark model at launch.
- Internal testing wraps end of June; a separate agentic Instagram shopping tool targets end of 2026.
- Frames Meta as following Anthropic and OpenAI into the consumer-agent race, not the chatbot race.
- Independent Rust-based coding agent for DeepSeek V4 Pro and V4 Flash; #2 on GitHub Trending today (+5,799 stars in 24 hours).
- Single binary, no Node/Python; ships an MCP client, sandbox, durable task queue, LSP diagnostics, and live cost tracking.
- Built around DeepSeek V4's 1M-token context and prefix cache, with native chain-of-thought streaming.
- v0.8.20 shipped today; 76 releases since launch from solo maintainer Hunter Bown.
- Markdown skills that encode senior-engineer workflows for AI coding agents — specs, tests, reviews, and ship gates.
- 20 skills across Define, Plan, Build, Verify, Review, and Ship phases; 3 specialist personas, 4 reference checklists, 7 slash commands.
- Works with any agent that accepts system prompts; MIT-licensed; baked-in practices from 'Software Engineering at Google'.
- +3,062 stars in the last 24 hours, sitting near the top of GitHub Trending.
- One tool call returns licensed financial datasets, real-time prices, and cited web sources — no per-vendor wiring.
- Covers filings, fundamentals, earnings, transcripts, estimates, ETFs, and market activity with inline citations on every result.
- Lands alongside Perplexity Computer for Professional Finance, which generates tearsheets, memos, and dashboards on top of the same data.
01
OpenAI Drops Three Realtime Voice Models With GPT-5-Class Reasoning
models openai.com
02
DeepMind's AlphaEvolve Posts Real Wins in Genomics, Power Grids, and Quantum
research deepmind.google
03
The Information: Meta Builds 'Hatch' Consumer Agent on Claude
industry theinformation.com
04
DeepSeek-TUI Crosses 20k Stars With a V4 Coding Agent in the Terminal
open-source github.com
05
Addy Osmani's Agent Skills Hits 33k Stars With 20 Production Workflows
tools github.com
06
Perplexity Adds Finance Search to the Agent API
tools perplexity.ai