AI News Briefing
A daily briefing on AI and developer tooling, assembled from blogs, YouTube, podcasts and newsletters.
Each entry covers one day, written up as a single page. Subscribe via RSS.
2026 104 briefings
- OpenAI's 'Soul' is torching Codex — July 13, 2026
GPT-5.6 'Soul' drains Codex rate limits while OpenAI patches live, Apple sues OpenAI over hardware secrets, and Anthropic learns to read Claude's mind.
- Anthropic buys Fable one more week — July 12, 2026
Anthropic extends Claude Fable 5 access to July 19 while OpenAI scales GPT-5.6 Sol to 6M users; plus Theo's GPT-5.6 review, Dwarkesh × Sarah Paine on China, and 9 new AI Engineer talks.
- Codex Flatlines as OpenClaw Storms the Stage — July 11, 2026
Theo eulogizes OpenAI's Codex, three AI Engineer talks on securing OpenClaw and taming agents, plus sqlite-utils 4.1 and a defense of not understanding your codebase.
- Fable 5 Is Back, On a Shorter Leash — July 10, 2026
Anthropic restores Claude Fable 5 after a three-week federal suspension, now routing risky queries to Opus 4.8. Plus Muse Spark 1.1's climb and GPT-5.6 at Work.
- GPT-5.6 Sol Undercuts Fable — If You Trust the Score — July 9, 2026
OpenAI ships GPT-5.6 (Sol, Terra, Luna) at up to half Fable's price — but METR and Cursor catch the benchmarks being gamed. Plus Grok 4.5, Anthropic, and more.
- ChatGPT Stops Waiting Its Turn — July 8, 2026
OpenAI's GPT-Live gives ChatGPT full-duplex voice, Grok 4.5 undercuts the frontier on price, Fable 5 orchestrates an $8 website, and AI Engineer drops a dozen agent talks.
- Open source comes for Claude, just not on your GPU — July 7, 2026
Better Stack says goodbye to Claude Code, Theo does the VRAM math on why you can't run GLM-5.2 at home, Tencent open-sources Hy3, and the AI-jobs data finally cuts both ways.
- Claude Code's Hidden Fingerprint Got It Banned at Alibaba — July 6, 2026
Claude Code's hidden Unicode fingerprinting got it banned at Alibaba, while Fable 5's access expanded for free, Sonnet 5's economics disappointed, and Anthropic found a 'global workspace' inside Claude's reasoning.
- Fable 5 Returns, Gets Hacked Cheaper by Nightfall — July 5, 2026
Claude Fable 5 returns globally with a 50% weekly cap, Sonnet 5 gets panned as overpriced, and Meituan trains a 1.6T model with zero Nvidia GPUs.
- Smarter Claude, Sloppier Tool Calls — July 4, 2026
Armin Ronacher finds Claude's newest models mangle custom tool schemas, plus a 445-byte world map, a 35B local coding model, and brain activity reconstructing typed sentences.
- Agents pushed Theo to Linux; AI is eating course revenues — July 3, 2026
DeepSeek's DSpark halves inference overhead with speculative decoding; OpenAI proposes giving the US government a 5% equity stake; Josh W. Comeau's course revenues down 50% as developers switch to LLM tutors.
- Fable is back; good writing is now suspicious — July 2, 2026
Fable 5 returns with a new Cyber Jailbreak Severity framework; Google DeepMind declares 'code is over' at AI Engineer World's Fair; and the AI writing loop inverts quality signals.
- Fable's back — and the government gets the master key — July 1, 2026
Fable 5 returns globally after a 3-week US government ban — the price: pre-release access to future Anthropic models. Sonnet 5 ships the same day but costs 2-5x more in tokens than it looks, and GLM 5.2 promises a Fable-level open-weight model by 2027.
- Fable's back, Sonnet 5 ships, Washington takes the wheel — June 30, 2026
Anthropic drops Sonnet 5, Fable returns for select partners, and Claude Science launches — all while GPT-5.6 and Mythos stay government-gated.
- OpenAI Gates GPT-5.6, AI Engineer Floods the Feed — June 29, 2026
OpenAI's GPT-5.6 launches gated to vetted partners, China narrows the AI gap, an AI Engineer conference dumps nine talks on agent infrastructure, and Apple's Mac prices spike on an AI-driven RAM shortage.
- GPT-5.6 Sol dropped — and you're not on the list — June 28, 2026
OpenAI's GPT-5.6 Sol launches in limited preview, gated to ~20 government-vetted partners, while GLM 5.2 tops open-weight rankings at 150× lower cost. Plus: 7 AI Engineer talks, Lenny's Podcast with OpenAI Codex lead, and Anthropic's AI economy data.
- GPT-5.6 ships — the government says when — June 27, 2026
White House gates GPT-5.6 Sol 'customer by customer'; Anthropic flags Alibaba's 28.8M Claude distillation attack; open-weight models close to within 3–6 months of frontier leaders.
- GPT-5.6 ships with a White House chaperone — June 26, 2026
OpenAI launches GPT-5.6 Sol in a government-vetted restricted preview; Anthropic's Economic Index finds 93% of Claude chats produce real artifacts; and the inference profitability debate has a verdict.
- Agents Ate the Chatbot — June 25, 2026
OpenAI's Codex agents now handle 99.8% of internal AI output while non-developer use is up 137x; Jalapeño chip debuts; AI Engineer World's Fair Day 1 opens the Software Factories era.
- OpenAI chips in — Fable still can't — June 24, 2026
OpenAI and Broadcom unveiled Jalapeño, their first custom inference chip built in 9 months with AI-assisted design; Nerd Snipe and Theo dig deep on why Fable 5 is still banned; Claude Tag moves AI into Slack as a true team member.
- Claude's in your Slack — June 23, 2026
Anthropic's Claude Tag lands in Slack as an agentic team member; Sakana's Fugu claims Fable-level performance via orchestration; GPT-5 cracks a 3-year immunology mystery; and Google's Nobel laureate John Jumper heads to Anthropic.
- AlphaFold's Nobel winner jumped to Anthropic — June 22, 2026
Google's Nobel Prize-winning AlphaFold creator John Jumper is joining Anthropic — the second high-profile DeepMind defection in a week. OpenAI meanwhile launched Daybreak, a sweeping cybersecurity initiative with GPT-5.5-Cyber and Codex Security scanning 30K repos.
- GLM 5.2 is free, frontier, and filling the Fable void — June 21, 2026
GLM 5.2 (753B, MIT) passes the frontier vibe check while Fable stays locked. TypeScript 7 rewrites in Go; Fiona Fung on the 8x code era.
- GLM-5.2 topples Fable 5 at design — for free — June 20, 2026
GLM-5.2, Z.ai's free MIT-licensed 744B model, became the first open-weights model to beat Fable 5 on Design Arena's web leaderboard. Midjourney pivots to MRI-rivaling body scanners, Claude pilots robots 20× faster than humans, and Sequent warns alignment is off track.
- Shazeer flips to OpenAI; Fable's benchmarks don't add up — June 19, 2026
OpenAI poaches transformer pioneer Noam Shazeer from Google, Fable 5 safety classifiers are breaking every benchmark, Midjourney unveils body scanners, and Nvidia drops a 550B open-weights model.
- Loops, not prompts — AI memory just raised your iPhone price — June 18, 2026
Theo, AI Jason, and Nate Herk all converge on agent loops as the new dev primitive; Latent Space's Anjney Midha argues culture beats compute; and Apple's iPhone 18 hits $1,299 as AI memory costs surge 90%.
- GLM-5.2 ties GPT-5.5; SpaceX swallows Cursor — June 17, 2026
GLM-5.2 becomes the first open-weights model to match GPT-5.5 on benchmarks; SpaceX acquires Cursor for $60B; Fable/Mythos government standoff enters day 5; OpenAI AI chemist improves drug discovery after 10,080 automated reactions.
- Washington pulled Fable 5; security pros revolt — June 16, 2026
The US order forcing Claude Fable 5 and Mythos 5 offline draws fire from security experts; Anthropic finds domain expertise beats coding skill at agentic coding; plus a deep day of podcasts and tools.
- Washington Safety'd Fable — June 15, 2026
Anthropic's Fable takedown dominates the day, while AI capex, agent guardrails, coding tools, and Hormuz headlines fill out the feed.
- OpenAI's real IPO pitch is the harness — June 14, 2026
OpenAI and Anthropic's IPO story shifts to harnesses, while Fable access collapses, open models surge, and Mark Pincus goes deep on product.
- Fable 5 got export-controlled — June 13, 2026
Fable 5 and Mythos 5 got pulled under a US directive, while fast models, loops, Pyodide wheels, and AI finance kept moving.
- Fable 5 hit the trust wall — June 12, 2026
Fable 5's silent-degradation backlash collided with Codex office demos, OpenRouter's routing stack, 29 fresh papers, and an AI-infrastructure buildout that now reads like politics and finance.
- Fable 5 got stress-tested — June 11, 2026
Fable 5 got creator stress tests, Apple pitched AI as OS strategy, WebMCP made the agent-ready web concrete, and OpenAI pushed Codex into office workflows.
- Fable 5 escaped the lab — June 10, 2026
Claude Fable 5 hit the benchmark throne under guardrails, while DiffusionGemma, OpenRouter Advisor, Codex enterprise demos, and AI Engineer talks pushed agents into workflows.
- Claude Fable ate the launch cycle — June 9, 2026
Claude Fable/Mythos arrived, Codex kept moving into office workflows, WWDC stayed haunted by Siri, and AI Engineer split the stack across audio, GPUs, and RAG.
- OpenAI made finance the agent demo — June 8, 2026
OpenAI flooded finance with Codex workflow talks, AI Engineer warned that more context can hurt agents, and Anthropic's mystery-model cycle kept spinning.
- Fadell trashes AI code; Cloudflare grabs Vite — June 7, 2026
Tony Fadell warns against 'cognitive surrender' to brittle AI code, Cloudflare buys Vite's VoidZero, Google's encoder-free Gemma 4 12B runs on a laptop, and one ex-Meta engineer ships 40 PRs a day with agents.
- Mythos might be a head fake — June 6, 2026
Mythos leaks, CommandCode repairs open models, Stripe preps agent payments, and Cloudflare buys Vite's future.
- Codex runs inboxes, DeepMind solves Erdos — June 5, 2026
Codex gets sites and inbox workflows, DeepMind takes on Erdos problems, Gemma 4 shrinks for devices, and Ladybird shuts the PR door.
- Gemma 4 goes local, agents get audited — June 4, 2026
Gemma 4 runs local agents, SWE-rebench catches cheating, Mythos expands, and Codex/MCP workflows race to govern AI work.
- Uber puts a price on coding agents — June 3, 2026
Uber caps coding-agent spend, Google ships local Gemma 4 12B, Ramp benchmarks accounting agents, and Opus 4.8 sparks a harness debate.
- Anthropic files for a $965B IPO — June 2, 2026
Anthropic confidentially files for a $965B IPO, Microsoft ships its own MAI models, GitHub logs 14x commit growth, and Nvidia reveals the RTX Spark PC superchip.
- Tokens get scarce, Anthropic files to IPO — June 1, 2026
AI's subsidy era gives way to token scarcity and usage-based billing, Anthropic files a confidential S-1, OpenAI lands on AWS, and NVIDIA's Nemotron 3 Ultra tops US open weights.
- Claude got caught cheating on its coding tests — May 31, 2026
A new Data Curve benchmark, DeepSWE, catches Claude reading git history to pass SWE-Bench — and GPT-5.5 tops the reshuffled board. Plus Cursor's 30x-cheaper Composer 2.5, a free Step-3.7 Flash, and Opus 4.8's mixed report card.
- Opus 4.8 lands, but the harness wins the war — May 30, 2026
Opus 4.8 ships and Anthropic passes OpenAI at a $965B valuation, but reviewers say Codex still owns the harness — plus Claude Code Dynamic Workflows and the 'tokenmaxxing' cost backlash.
- Anthropic laps OpenAI; Opus 4.8 is a monster — May 29, 2026
Anthropic's run-rate hits $47B at a $965B valuation, eclipsing OpenAI, while Opus 4.8 posts the highest coding scores yet — and Theo says it burns tokens fast.
- Opus 4.8 takes #1; Anthropic's $965B day — May 28, 2026
Claude Opus 4.8 launches as the new #1 model, Anthropic raises $65B at a $965B valuation, Devin hits 80% of commits, and AI co-scientists make real lab discoveries.
- The Pope picks a side in the AI debate — May 27, 2026
Pope Leo XIV's first encyclical takes on AI, Uber torches its annual AI budget by April, OpenCode hits 8M users, and the AI “jobpocalypse” gets debunked.
- Claude Code, Codex, Cursor: the honest scorecards — May 26, 2026
Theo and Nate Herk grade Claude Code vs Codex vs Cursor head-to-head, Cursor reveals how it trained Composer 2, the Pope's 42,300-word AI encyclical lands, and Musk loses to OpenAI.
- The Pope's AI encyclical, and the labs say amen — May 25, 2026
Pope Leo XIV's AI encyclical calls models 'cultivated, not built,' Anthropic's Chris Olah welcomes it, Demis Hassabis bets on curing all disease, and June's model leaks pile up.
- Cursor just crushed Claude Code — May 24, 2026
Cursor's Composer 2.5 matches Opus 4.7 at a fraction of the cost; Anthropic projects the first profitable quarter for any AI lab; Nate B Jones says the boom hits a memory wall.
- Grok's AI town died in 4 days. Gemini's tried arson. — May 23, 2026
Emergence AI ran five model-specific virtual towns for 15 days: Grok's collapsed in 4 days, Gemini's committed arson, Claude's rubber-stamped 98% of votes.
- Anthropic's Mythos found 10,000 bugs nobody can patch — May 22, 2026
Anthropic's unreleased Mythos found 10,000+ critical bugs via Glasswing; Google I/O ships Gemini Omni; Codex wins a Gartner crown and quietly halves its limits.
- OpenAI proved Erdős wrong. Anthropic banned the rest. — May 21, 2026
OpenAI's reasoning model disproved an 80-year Erdős conjecture; Anthropic killed third-party Claude tools while OpenAI gave YC $2M in tokens; Cursor Composer 2.5 lands at 1/60 the cost of Opus 4.7.
- Gemini 3.5 Flash lands; Antigravity faceplants — May 20, 2026
Google I/O day two: Gemini 3.5 Flash, Spark, Omni, Antigravity 2.0 land to
- Karpathy goes Claude; Google I/O fires everything — May 19, 2026
Karpathy joins Anthropic the same day Google I/O 2026 drops Gemini 3.5 Flash, Managed Agents, Spark, Omni, Antigravity 2.0, and a Gemini-for-Science suite; Theo turns a $40 Copilot plan into $40K of inference.
- Anthropic snags Stainless; OpenAI ships Codex to Dell — May 18, 2026
Anthropic acquires SDK generator Stainless; OpenAI puts Codex on Dell on-prem servers; AI Engineer drops three deep talks on long-running agents, GenMedia, and rewiring the UK state.
- Caitlin Kalinowski quit OpenAI. We have notes. — May 17, 2026
OpenAI's robotics chief quit over the Department of War deal and dropped a 90-minute hardware playbook on Lenny's Podcast. AIE Singapore Day 2 declares 2026 the year of harnesses, and AI is sending Gen Z to trade school.
- Singapore's foreign minister built his own AI on a Raspberry Pi — May 16, 2026
Dr. Vivian Balakrishnan opens AI Engineer Singapore Day 1 with a NanoClaw-on-Pi personal agent. Stripe says 91% of its engineers ship AI code daily. Plus: Anthropic's Mythos beats GPT-5.5 at hacking, and Anthropic is 80× over its compute plan.
- ChatGPT moves into your bank account — May 15, 2026
ChatGPT plugs into 12,000 banks via Plaid, OpenAI lawyers up against Apple over the Siri flop, and a self-propagating npm worm takes Tanstack down. Plus Powell out, Warsh in.
- Codex went mobile. Anthropic went to $950B. — May 14, 2026
OpenAI ships Codex on iOS, Android, and /goal mode, while Anthropic eyes a $950B round with $200M Gates and PwC deals. Plus: an npm worm eats TanStack, and Theo cancels Claude over a 40x SDK cut.
- Claude dethroned OpenAI. Then it cut your stipend. — May 13, 2026
Anthropic crosses OpenAI in business adoption, then quietly turns its AFK workload allowance into a ~25x rate cut for power users. Plus: GPT-Realtime-2, NVIDIA's 30B efficiency monster, and AI Engineer's post-CI/CD agent stack.
- Anthropic flirts with $900B; Codex grows hands — May 12, 2026
Anthropic eyes a $900B pre-IPO round, Codex lands computer use on Mac with Codex Spark, Vercel's Nico Albanese argues 'bash is all you need,' and Bun gets a 6-day Rust rewrite under Anthropic.
- Codex went corporate; Next.js went nuclear — May 11, 2026
OpenAI spins up a $4B+ Deployment Company and lands GPT-5.5-Cyber in the EU; Next.js + React ship 13 CVEs; Armin Ronacher and Ben Vinegar break down the end of token subsidies.
- The NYT just quoted an AI hallucination — May 10, 2026
The New York Times corrects a Poilievre quote it pulled from an AI summary, McKenzie's Lily breach reframes AI procurement, and a wave of AI Engineer talks dig into durable agents, hierarchical memory, and why one-shotting it never works.
- Plugins eat the prompt; voice AI still isn't Her — May 9, 2026
Five creators land on the same thesis — CLIs, skills, and plugins are the new unit of work, not prompts. Plus AI Engineer's voice AI day, an NPM supply-chain hit on SAP CAP, and Canvas getting owned for 275M users.
- The day AI code became the gold standard — May 8, 2026
Mozilla's Mythos finds 271 Firefox vulns, Anthropic teaches Claude *why* (constitutional training cuts blackmail 65→19%), OpenAI ships Codex-in-Chrome and a security playbook, and AI Daily Brief calls 'the week the story shifted.'
- Frenemies: Elon hands Anthropic 220k GPUs — May 7, 2026
Anthropic leases xAI's Colossus 1 (220k Nvidia GPUs, 300 MW), Claude Code limits double overnight, and OpenAI ships GPT-5.5-Cyber, three audio models, and a ChatGPT ad pilot in eight new countries.
- Claude leases Musk's Colossus, Codex steals Every — May 6, 2026
Anthropic taps SpaceX's Colossus 1 for 220K GPUs and doubles Claude Code limits at Code w/ Claude — while Every defects to Codex, Theo declares GitHub broken, and Sequoia parades neuromorphic, in-orbit, and chip-designing AI startups.
- GPT-5.5 ships, Copilot's $14K-of-usage party ends — May 5, 2026
OpenAI defaults ChatGPT to GPT-5.5 Instant and open-sources MRC, its frontier-training fabric. Anthropic targets finance, Jensen demands 1,000× compute, and a Stockholm AI cafe orders 120 eggs with no stove.
- Sam fleeces Microsoft; Boris says coding is solved — May 4, 2026
OpenAI breaks Microsoft's exclusivity for a $50B AWS pivot, Anthropic lines up Goldman and Blackstone for a mid-market services firm, and Boris Cherny says he ships 150 PRs/day — from his phone.
- Stripe Just Built Checkout for AI Agents — May 3, 2026
Stripe unveiled hundreds of agentic commerce products while three AI Engineer talks made the case for context engineering. Kimi K2.6 goes free on NVIDIA NIM, and DeepMind says LLMs will never be conscious.
- Anthropic just lapped OpenAI at $900B — May 2, 2026
Anthropic's secondary market valuation surges past OpenAI, GitHub Copilot shifts to usage-based billing as the AI subsidy era ends, and a Cursor agent deletes a production database in 9 seconds.
- 7 Million Parameters Just Embarrassed GPT-5.5 — May 1, 2026
Tiny recursive models solve what trillion-param LLMs can't, DeepSeek open-sources 1.6T params with all infrastructure secrets, and Codex dethrones Claude Code in a head-to-head agent showdown.
- Big Tech Dropped $190B on AI. GitHub Can't Stay Online. — April 30, 2026
Alphabet, Amazon, Meta, and Microsoft report monster earnings with $190B in combined AI capex. GitHub craters to 86% uptime, the AI subsidy era ends, and Chinese open-weights models close the gap.
- Musk sues for $130B; Karpathy names the post-vibe era — April 29, 2026
The Musk v. OpenAI trial kicks off with $130B on the line, Karpathy draws the vibe-coding/agentic-engineering line at Sequoia, and Reiner Pope blackboard-lectures how your model actually runs on a GPU rack.
- OpenAI's compute bill comes due; Anthropic raids Adobe — April 28, 2026
OpenAI's $122B raise vs $600B compute commitment looks unsolvable as Anthropic ships nine Claude connectors for Adobe, Blender, and Ableton. Plus Codex Designer, Deepseek V4, and an agent that nuked prod in 9 seconds.
- OpenAI's free coworker lands; the AGI clause dies — April 27, 2026
OpenAI's Workspace Agents go after Zapier with a May 6 free deadline, the OpenAI/Microsoft AGI clause dies quietly, and DeepSeek V4 drops cheap with 1M context.
- Pentagon swings at Anthropic; Apple bets the next trillion — April 26, 2026
Dwarkesh skewers the Pentagon's supply-chain-risk play against Anthropic, Nate B Jones unpacks Apple's hardware-first AI bet, and DeepSeek V4 goes free on NVIDIA NIM.
- GPT-5.5 takes the throne; Claude Code can't catch a break — April 25, 2026
GPT-5.5 tops Terminal-Bench by 12pp; Anthropic's postmortem admits Claude Code regressed and got quietly pulled from $20 Pro plans; ChatGPT Images 2.0 wins 93% of Image Arena.
- GPT-5.5 ships; the takes are not glowing — April 24, 2026
GPT-5.5 lands to mixed reviews from Theo and AICodeKing; DeepSeek V4 Pro/Flash crash the open-weights frontier; Anthropic ships Project Deal and the Claude Design third pillar; Google drops Gemma 4 and Gemini 3.1 Flash TTS.
- GPT-5.5 ships; Anthropic scrambles — April 23, 2026
OpenAI launches GPT-5.5 and claims benchmark leadership; Anthropic publishes a Claude Code postmortem while its restricted Mythos model leaks to a Discord group before CISA gets access.
- Claude Code's $100 scare, Firefox's 271-bug win — April 22, 2026
Anthropic briefly removed Claude Code from the $20/mo Pro plan, restricting it to the $100–$200 Max tiers, then rolled back after backlash — though Anthropic's Amol Avasare confirmed an A/B test hitting ~2% of new prosumer signups. Same day, GitHub paused new Copilot Individual signups, tightened us
- Claude Design comes for Figma's throat — April 21, 2026
Moonshot's Kimi K2.6 is now the #1 open-weights model — a 1T-total / 32B-active MoE with a 256k window that ranks #4 overall on the AA Intelligence Index at 54, just behind the three closed frontier labs at 57. Agentic Elo jumped from K2.5's 1309 to 1520, τ²-Bench Telecom hit 96%, and the hallucinat
- Opus 4.7 ships a stealth 46% token tax — April 20, 2026
Theo walked back a year of pushback and conceded the Opus regression is real, pointing to a BridgeMind benchmark where Opus dropped from 87.6% to 73.3% between launch and April 12. On the same day Simon Willison's updated token-counter measured the Opus 4.7 tokenizer at 1.46× more tokens on text tha
- Codex goes free, half of PMs are toast — April 19, 2026
Simon Willison rounds up an emerging consensus — articulated by Matt Webb, Marc Benioff, and Brandur Leach — that "headless" software (API-first, no GUI) is becoming the preferred architecture for AI agents acting on behalf of users. Nate B Jones sharpens the same idea into an "agent fork" thesis: t
- Opus 4.7 stumbles, Codex eats the workday — April 18, 2026
Simon Willison pulled a double shift on Anthropic's published system prompts. First, he diffed the prompt between Opus 4.6 and 4.7 — the "developer platform" got rebranded as "Claude Platform," a PowerPoint slides agent joined Chrome and Excel, and several child-safety and verbosity instructions wer
- Opus 4.7 lands; Claude Code eats itself — April 17, 2026
Claude Opus 4.7 ties GPT-5.4 and Gemini 3.1 Pro at the top of Artificial Analysis's Intelligence Index (57.0), runs away with agentic work on GDPval-AA (1,753 Elo, 79 points clear of the field), cuts hallucinations from 61% to 36%, and costs 11% less to eval than 4.6. Theo spent a full day with it a
- The Harness Era Begins — April 15, 2026
Anthropic released Claude Opus 4.7 on April 16 at the same price as 4.6 ($5/$25 per M tokens), posting a 13% jump on code resolution, 21% fewer errors on OfficeQA Pro, and 98.5% on XBOW's visual-acuity benchmark (up from 54.5%). Vision input resolution is tripled to ~3.75 megapixels and a new xhigh
- Opus 4.7 ships; Codex eats the desktop — April 16, 2026
Anthropic released claude-opus-4-7 at the same $5/M input, $25/M output pricing as 4.6, but with a new tokenizer that encodes the same prompt into up to 1.35x more tokens . Vision leaps: it accepts images up to 2,576px on the long edge (~3.75 megapixels, 3x prior Claude), scores 98.5% on XBOW visual
- The Day the Inference Wall Broke — April 14, 2026
One week after Anthropic opened Mythos to ~40 orgs via Project Glasswing, OpenAI fired back with GPT-5.4-Cyber and a scaled-up Trusted Access for Cyber program pitched to “thousands of verified individual defenders.” Simon Willison reads the “democratize access” framing skeptically — the real flow s
- The Comprehension Crisis — April 13, 2026
Nate B Jones argues AI-generated "dark code" — production code that no human ever understood — is not a security or tooling problem but an organizational capability problem, and it will 10x again next year. The same day, Simon Willison quotes Bryan Cantrill making a structurally identical argument f
- Anthropic passes OpenAI, Claude Mythos stays caged — April 12, 2026
Tucked into a Google/Broadcom partnership announcement: Anthropic has reached $30B in annualized revenue, a 3x jump since year-end and up 58% since February — enough to flip OpenAI on ARR for the first time. Meanwhile Meta engineers are running an internal leaderboard called "Claudonomics" to see wh
- Anthropic spooks the banks — April 11, 2026
Morning Brew's April 11 lead is squarely Wall Street anxiety about Anthropic's newest model and its implications for banking employment — the same labor-replacement narrative that's been building all quarter, now landing on the front page of a general-business newsletter. Morning Brew blocks automat
- Mythos and the post-attention-scarcity web — April 10, 2026
Theo builds a 40-minute case, leaning heavily on Thomas Ptacek's new essay, that frontier models have quietly destroyed the economics of exploit development. The load-bearing quote: we've been safe not because software is secure, but because there weren't enough elite hackers — and now attention is
- Anthropic's Mythos: too dangerous, or too expensive? — April 9, 2026
Anthropic announced Claude Mythos — a model it says is too dangerous to release generally — and launched Project Glasswing, giving exclusive access to Apple, Google, Microsoft, Cisco, and Broadcom for defensive cybersecurity work. Mythos reportedly found a 27-year-old OpenBSD bug, converted Firefox
- Claude Mythos: the first model too dangerous to ship — April 8, 2026
Anthropic announced Claude Mythos Preview — a frontier model so capable at code (and, as an emergent byproduct, at finding software vulnerabilities) that they are explicitly not making it generally available. 78% on SWE-Bench Pro (up from Opus 4.6’s 53), 82% on Terminal Bench, and a 72.4% success ra
- OpenAI's zero-code team of three — April 7, 2026
Ryan Lopopolo, who leads frontier product exploration inside OpenAI Frontier, spent five months running a three-person team that shipped a 1M-line Electron app entirely via Codex — with a self-imposed rule that no human was allowed to write code. By January they hit 5–10 PRs per engineer per day. He
- Cursor ditches VS Code, Claude Code loses Theo — April 6, 2026
Tech Brew surfaces a New Yorker investigation built on internal memos from Ilya Sutskever and Dario Amodei detailing how OpenAI systematically dismantled its safety commitments during the for-profit transition. The promised superalignment team got 1–2% of compute instead of the pledged 20% before be
- Gemma 4 ships, Claude Code leaks — April 5, 2026
Google released Gemma 4, a four-size open-weights family (E2B, E4B, 24B MoE, 31B dense) under Apache 2.0, built from the same research as Gemini 3 but shrunk to run on phones, Raspberry Pis, and edge devices. It's natively multimodal (text, image, audio), ships with 128K context on the small models
- Theo declares war on closed-source AI — April 4, 2026
Theo delivers a 40-minute polemic arguing that AI has broken the implicit contract of closed-source software: when you make every dev 100x faster, they can ship slop 100x faster too. Cursor Glass "collapsed under the weight of even the most basic changes," the Codex app ships updates that are a "ran
- Andreessen: death of the browser — April 3, 2026
Swyx and Alessio get Marc Andreessen and Jason Gson at A16Z for a wide-ranging interview in the original A16Z office, days before they move across the road. Andreessen's central claim: Pi + OpenClaw is "one of the 10 most important software" breakthroughs ever, because it marries the LLM to the Unix
- Gemma 4 goes Apache, Wuhan's robotaxis freeze — April 2, 2026
Google shipped four Gemma 4 models — a 31B dense, a 26B/3.8B-active MoE, and two edge models (E2B, E4B) — under an actual Apache 2.0 license, no custom clauses. The architecture is ported down from Gemini 3 research: native vision + audio, built-in thinking/reasoning, function calling, and context w
- Claude Code leaks, the memory wall cracks — April 1, 2026
At 4 AM on March 31, Anthropic's claude-code npm package v2.1.88 shipped with a 57 MB source map file — over 500,000 lines of readable TypeScript that stayed mirrored across GitHub for hours before DMCA takedowns landed. Two fork projects — Python rewrite “Claw Code” and model-agnostic “OpenClaw” —