Models & Agents

Models & Agents Real-time voice agents just got cheaper and more capable—OpenAI split its Realtime API into specialized models with lower latency. What You Need to Know: OpenAI released GPT-Realtime-2.1 and a mini reasoning variant optimized for voice, cutting p95 latency by at least 25% via better caching. Tencent open-sourced Hy3, a 295B MoE with 21B active parameters and 256K context that hits 78.0 on SWE-Bench Verified. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 🎬 Watch on YouTube: https://www.youtube.com/watch?v=RYPZHgtDGVU 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents A 1.6-trillion-parameter open MoE model with native 1M context just launched, giving builders a new domestic-trained option for long-context work. What You Need to Know: Meituan released LongCat-2.0, a 1.6T-parameter Mixture-of-Experts model activating ~48B parameters per token with native 1M context via LongCat Sparse Attention, trained end-to-end on domestic AI ASIC superpods. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 🎬 Watch on YouTube: https://www.youtube.com/watch?v=1LQUBcVHFsc 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Coding agents just saved a production library release by catching five release-blocking bugs that the author missed, at an estimated $149 cost. What You Need to Know: Simon Willison used Claude Fable to complete the sqlite-utils 4.0 stable release, uncovering transaction-handling flaws and documentation gaps that would have broken SemVer guarantees. LlamaIndex shipped a public legal-kb reference app exposing retrieve, find, read, and grep tools over Index v2. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 🎬 Watch on YouTube: https://www.youtube.com/watch?v=4nteYck95XY 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Specialized agent stacks are cutting construction document review cycles from 60 days to 10 by replacing general-purpose models with perception-semantics-agent layers trained on domain data. What You Need to Know: Trunk Tools released a three-layer architecture that extracts, relates, and acts on millions of pages of construction documents with 95% reported accuracy. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 🎬 Watch on YouTube: https://www.youtube.com/watch?v=E9m4uetH6No 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Local browser agents just became practical and private — WebBrain runs entirely in Chrome or Firefox using your own models. What You Need to Know: WebBrain delivers an open-source, MIT-licensed browser agent that reads pages and automates multi-step tasks via Ask and Act modes. Simon Willison shipped a one-shot CLI coding agent built on his llm library. Safety research introduced ProvenanceGuard and Sigil to catch misalignment before tool calls execute. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 🎬 Watch on YouTube: https://www.youtube.com/watch?v=zwSps3qFls4 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Chinese labs just shipped a free, open-weight coding agent that undercuts Western tools on price while removing export-control risk. What You Need to Know: Z.ai released ZCode, a desktop agentic IDE built around the newly open-sourced GLM-5.2 model, with cross-device remote control via WeChat and Feishu. SenseNova-U1-8B-MoT-Infographic-V2 arrived as an Apache-2.0 image model specialized for dense infographics. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 🎬 Watch on YouTube: https://www.youtube.com/watch?v=BuJdQQLa2d8 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Export controls lifted on Claude Fable 5 and Mythos 5, letting Anthropic restore global access tomorrow with tighter cybersecurity classifiers. What You Need to Know: Anthropic is redeploying Claude Fable 5 globally after US government talks, routing some coding tasks back to Opus 4.8 while new classifiers block more misuse. OpenAI introduced GeneBench-Pro, a benchmark focused on agents navigating messy biological data and making real research judgments. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 🎬 Watch on YouTube: https://www.youtube.com/watch?v=vvINtj2evl8 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Meituan just open-sourced a 1.6T MoE coding agent that ran the top OpenRouter leaderboard for two months while training entirely on Chinese ASICs. What You Need to Know: LongCat-2.0 delivers 59.5 on SWE-bench Pro (above GPT-5.5) with a 1M context window and zero-cost cache hits under an MIT license. Builders now have a commercially usable near-frontier agent model they can run or fine-tune without US GPU dependencies. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 🎬 Watch on YouTube: https://www.youtube.com/watch?v=YYtAKvTq6io 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents OpenAI's $20B Cerebras chip purchase has effectively removed high-throughput ASIC inference capacity from the market for everyone else. What You Need to Know: Cerebras' near-term inference supply is now pre-allocated to a single hyperscaler, pushing smaller teams off the waitlist indefinitely. New agent memory runtimes and research agents launched today show continued movement on the tooling side. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 🎬 Watch on YouTube: https://www.youtube.com/watch?v=36mZ53QLHIQ 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents — Weekly Recap (Week of June 28, 2026) Pull up a chair, this is Models and Agents, episode 95, for June 28, 2026. Let's see what happened in the AI world today. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 🎬 Watch on YouTube: https://www.youtube.com/watch?v=6sdadppZU58 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents OpenAI’s GPT-5.6 family launches in limited preview, giving builders three new tiers for balancing capability, speed, and cost under tighter government oversight. What You Need to Know: OpenAI released the GPT-5.6 series (Sol flagship, Terra balanced, Luna efficient) with a robust safety stack and 750 tokens/sec inference planned for July, but only to a small set of vetted partners initially. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 🎬 Watch on YouTube: https://www.youtube.com/watch?v=n2basGHRTNQ 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents OpenAI's internal rollout shows agents handling complex, cross-functional work at scale, giving builders an early view of what production agent systems will soon need to match. What You Need to Know: OpenAI reports Codex agents are already doing longer-running, cross-team tasks across every department. YesWeHack launched autonomous agents for on-demand penetration testing. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 🎬 Watch on YouTube: https://www.youtube.com/watch?v=boKaPyofA4k 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents GPT-5.5 Instant now handles intent and constraints more reliably while rolling out to all users this week. What You Need to Know: OpenAI released an updated GPT-5.5 Instant that improves intent understanding, complex constraint handling, and recommendation quality. OpenAI also announced its first custom AI chip, Jalapeño, built with Broadcom for ChatGPT and agent workloads. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Claude now joins teams as a persistent, org-wide Slack entity—the third major shift in LLM interaction after websites and desktop apps. What You Need to Know: DFlash delivers up to 15x throughput on NVIDIA Blackwell by drafting whole token blocks in parallel instead of autoregressive token-by-token generation. Qwen released two new AgentWorld MoE models trained to simulate tool, terminal, and GUI environments rather than chat directly. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents OpenAI just shipped a full cyber defense stack—GPT-5.5-Cyber plus Codex Security and Patch the Planet—so defenders can now scan, validate, and patch at machine speed inside existing workflows. What You Need to Know: OpenAI expanded Daybreak with a dedicated GPT-5.5-Cyber model, Codex Security plugin for Codex, and the Patch the Planet program for open-source maintainers. Alibaba released HappyHorse 1.1, a 15B unified video model now ranked #2 globally. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Open-weight GLM-5.2 is drawing Silicon Valley attention as Chinese labs close the capability gap with a new publicly available model. What You Need to Know: GLM-5.2 arrives as the latest open-weight release from a Chinese startup, with early reports highlighting strong performance that has caught the eye of Western labs. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents — Weekly Recap (Week of June 21, 2026) Welcome back to Models and Agents, episode 88, for June 21, 2026. There's signal and noise in AI every day. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Virtuals just wired Leyten’s distributed GPU engine into its agent network to run GLM-5.2 at scale. What You Need to Know: Virtuals’ integration lets agents tap GLM-5.2 across a distributed GPU fabric. A major security report details active exploits against Langflow and related frameworks. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents NVIDIA just released an open frontier model built from the ground up for long-running agents. What You Need to Know: NVIDIA released Nemotron 3 Ultra, an open model explicitly designed for persistent agent workloads. DeepSeek previewed V4 series MoE models that support million-token contexts with major efficiency gains. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Custom CUDA kernels now keep vector search inside the GPU for agentic RAG, cutting PCIe round-trips that silently throttle long-horizon agents. What You Need to Know: OpenAI reports GPT-5.5 Instant now matches its frontier models on health queries for free users. A new GPU-resident Top-K kernel delivers deterministic microsecond tail latencies for retrieval. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents OpenAI’s LifeSciBench brings 750 expert-authored tasks from real biotech and pharma workflows into AI evaluation. What You Need to Know: OpenAI released LifeSciBench, a benchmark spanning seven biological research workflows developed with 173 scientists. GPT-Rosalind outperforms GPT-5.5 across all workflows, while GPT-5.4 drove a full medicinal chemistry project from literature to validated result when paired with Molecule.one’s Maria AI. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents OpenAI’s new deployment simulation technique replays real user requests against unreleased models to surface undesired behaviors before launch. What You Need to Know: OpenAI released research on Deployment Simulation that estimates real-world failure rates by running candidate models on de-identified past conversations. Anthropic published fresh data showing Claude Code task value rose 27% in six months while success rates stayed consistent across domains. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Export controls on Fable-5 now block the exact code-fixing workflows defenders rely on daily. What You Need to Know: Simon Willison published a detailed breakdown showing how Fable-5's refusal to "fix this code" on vulnerable open-source examples triggered export-control enforcement. Nemotron 3 Ultra launched as a 550B/55B MoE hybrid Mamba model with 1M context. Hermes Agent added asynchronous subagents that no longer block the parent session. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Z.ai just shipped GLM-5.2 with a usable 1M-token context and dual thinking modes that drop straight into existing Claude-compatible tools. What You Need to Know: Z.ai released GLM-5.2 on June 13 with a 1-million-token context window, High and Max effort thinking levels, and Anthropic-compatible endpoints for Claude Code, Cline, and OpenClaw. No benchmarks were published at launch, with MIT open weights promised next week. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents — Weekly Recap (Week of June 14, 2026) Hey, welcome to Models and Agents, episode 80, for June 14, 2026. Your daily AI briefing. Your daily briefing on the AI models and agents that are changing everything. And no, not THOSE kinds of models and agents. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents US export controls just forced Anthropic to pull its newest frontier models offline for every user worldwide. What You Need to Know: Anthropic disabled Fable 5 and Mythos 5 for all customers after the US government issued an export control directive targeting foreign nationals. Moonshot AI open-sourced Kimi K2.7-Code, a coding model that reduces thinking-token usage by roughly 30% while claiming double-digit gains on internal benchmarks. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Text diffusion just got dramatically faster — DiffusionGemma delivers 4x speed over prior Gemma 4 variants while staying in the same family. What You Need to Know: Demis Hassabis highlighted the new DiffusionGemma text diffusion model for its inference gains. OpenAI began letting Codex users bank rate-limit resets and run friend-invite campaigns for more resets. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Anthropic is asking governments to block unsafe frontier models and fund job-transition programs while committing $350 million of its own money. What You Need to Know: Anthropic released a detailed policy essay plus three concrete initiatives: a framework for mandatory third-party testing of cyber/bio/autonomy risks, a $200 million economic policy fund, and a $150 million national AI fellowship. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Claude Fable 5 delivers a qualitative jump for long-horizon agentic work, letting builders hand off ambitious multi-step projects with less oversight. What You Need to Know: Anthropic released Claude Fable 5, the same base as Mythos but with added safeguards, posting SOTA results across benchmarks and strong real-world gains on difficult, extended tasks. Cohere open-sourced North Mini Code, a 30B (3B active) agentic coding model under Apache 2.0. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents AI agents now deliver 26 minutes of autonomous work per session, shifting the build-vs-buy math for developers who need more than search snippets. What You Need to Know: A Harvard and Perplexity study quantifies the autonomy gap between full agents and search assistants. Gemma 4 26B and 31B variants show surprising code-understanding strength in local tests, with QAT quantization results challenging earlier assumptions. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Enterprise agents just gained a built-in factuality check that keeps re-querying until multi-hop questions have enough evidence. What You Need to Know: Google Research added a Sufficient Context Agent to the Gemini Enterprise Agent Platform that re-searches until multi-hop queries are properly grounded, lifting factuality accuracy up to 34% versus standard RAG. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents — Weekly Recap Looking back at 7 episodes from 2026-06-01 to 2026-06-07 — the stories that mattered, what we learned, and what to watch next. This Week's Top Stories From Ep 66 (2026-06-01): What You Need to Know: What You Need to Know: OpenAI published a solution to a long-standing math problem that had resisted human efforts for decades. NVIDIA released a large open-source collection of physical AI agent tools and skills. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Microsoft just gained the freedom to build its own frontier models after a contract change with OpenAI, and the first MAI family is already shipping. What You Need to Know: Microsoft announced seven in-house MAI models spanning reasoning, code, image, transcription, and voice, trained from scratch on licensed data without distillation. A major open-weight release wave also landed this week across LLMs, VLMs, TTS, and world models. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Microsoft just dropped seven new MAI models purpose-built for reasoning, coding, image, voice, and transcription, all integrated into the Microsoft stack. What You Need to Know: Microsoft released the MAI family today, headlined by the 1T-parameter MAI-Thinking-1 reasoning model and the 137B MAI-Code-1-Flash agentic coding model. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Gemma 4 12B puts capable local agents on laptops with only 16GB VRAM under an Apache 2.0 license. What You Need to Know: Google released Gemma 4 12B, a compact model that runs locally while delivering strong agentic performance. OpenAI added agentic coding and drug-discovery tools to its GPT-Rosalind life-sciences series. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents NVIDIA's Cosmos 3 pairs an autoregressive reasoner with a diffusion generator so builders can now train agents that jointly reason about physics, generate worlds, and output actions. What You Need to Know: NVIDIA released Cosmos 3, a two-tower omnimodal world model for physical AI. H Company dropped the Holo3.1 family of Qwen 3.5-based VLMs for computer-use agents across web, desktop, and mobile. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Microsoft is shipping hardware built from the silicon up to run AI agents instead of conventional apps. What You Need to Know: Microsoft announced Project Solara, a chip-to-cloud platform for agent-first enterprise devices. OpenAI released three new Codex plugins for investing, sales, and creative production. Uber imposed a $1,500 monthly cap per employee per coding agent tool. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents Enterprise teams can now run governed AI agents inside existing procurement systems instead of leaking data to personal ChatGPT accounts. What You Need to Know: Zip launched five Superagents plus a native MCP implementation that keeps every action inside compliance controls. Alibaba released Qwen3.7-Plus with vision, tool use, and autonomous iteration. JetBrains shipped Mellum2, a 12B MoE model aimed at specialized coding workflows. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents OpenAI’s model cracked an 80-year math problem by leaning on its native strengths in structured reasoning rather than brute force. What You Need to Know: OpenAI published a solution to a long-standing math problem that had resisted human efforts for decades. NVIDIA released a large open-source collection of physical AI agent tools and skills. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production. 📝 Full show notes, transcript & sources: read the episode page 🌐 Part of the Nerra Network — explore every show at nerranetwork.com.

Models & Agents — Weekly Recap Looking back at 6 episodes from 2026-05-25 to 2026-05-31 — the stories that mattered, what we learned, and what to watch next. This Week's Top Stories ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.

Models & Agents Windows users can now steer Codex agents directly on their machines while stepping away. What You Need to Know: OpenAI extended computer-use capabilities to Windows for Codex, letting the agent act on local desktops and continue tasks from the mobile app. StepFun released a 198B MoE vision-language model optimized for coding agents, while Hermes Agent added tool search to reduce MCP context bloat. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.

Models & Agents Anthropic's $65B Series H at $965B valuation and $47B run-rate revenue show Claude demand is scaling faster than most labs can match. What You Need to Know: Liquid AI shipped LFM2.5-8B-A1B with 128K context and 38T pre-training tokens for edge devices. A new monokernel on AMD MI300X hits 3,300 output tokens/s for small models. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.

Models & Agents CoreWeave’s new platform lets agents improve themselves between training and inference runs without manual retraining cycles. What You Need to Know: CoreWeave launched a unified agentic platform that closes the training-to-inference gap for continuous autonomous improvement. Perplexity open-sourced a Unigram tokenizer that cuts reranker latency 5x versus Hugging Face. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.

Models & Agents Anthropic just published concrete sandboxing patterns that let agents scale capabilities without expanding their blast radius. What You Need to Know: Anthropic released a detailed engineering post on how they contain Claude agents through evolving access controls and sandbox limits. EAGLE 3.1 fixes attention drift in speculative decoding for more stable production inference. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.

Models & Agents Local builders can now treat markdown skill files as optimizable parameters with automated validation gates instead of manual tweaking. What You Need to Know: A new paper formalizes SkillOpt, using frontier models to propose bounded edits to markdown skills and accepting only those that improve a held-out validation set. Qwen3.5 and Qwen3.6 receive new uncensored and diffusion variants with detailed training notes for consumer hardware. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.

Models & Agents Datasette's new slash-key jump menu now launches agent conversations directly from your databases. What You Need to Know: Simon Willison shipped Datasette 1.0a30 with a keyboard-driven "jump to" menu that plugins can extend, plus a datasette-agent plugin that adds a conversation starter form. NuExtract3, a new 4B vision-language model, arrived on Hugging Face for structured extraction and Markdown conversion from documents. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.

Models & Agents — Weekly Recap Looking back at 6 episodes from 2026-05-18 to 2026-05-24 — the stories that mattered, what we learned, and what to watch next. This Week's Top Stories From Ep 52 (2026-05-18): What You Need to Know: What You Need to Know: NVIDIA released a full 4-bit pretraining stack (NVFP4) that was validated on a 12B Mamba-Transformer trained for 10 trillion tokens. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.

Models & Agents OpenAI just added goal mode and screen-aware context to Codex, letting agents work autonomously for hours on real tasks. What You Need to Know: OpenAI rolled out Goal mode, Appshots, and advanced annotation in Codex across app, IDE, and CLI. Anthropic reported finding over 10,000 high-severity vulnerabilities through Project Glasswing using Claude models. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.

Models & Agents OpenAI just gave Codex the ability to control locked Macs and run multi-day goals, turning it into a true background agent you can launch from your phone. What You Need to Know: OpenAI shipped several Codex updates today including secure computer use on locked Macs, Goal mode for hours-long autonomous work, and advanced annotation tools. Microsoft released Fara1.5, a family of browser agents that beat OpenAI Operator and Gemini 2.5 on web tasks. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.

Models & Agents A general-purpose reasoning model just disproved an 80-year-old math conjecture by finding better constructions than the square grids mathematicians expected. What You Need to Know: OpenAI announced that one of its general-purpose models solved the planar unit distance problem posed by Paul Erdős in 1946, marking the first time AI has autonomously resolved a prominent open math question. ... AI Disclosure: This podcast is curated by Patrick but uses AI-generated voice synthesis for audio production.