More than 1,100 employees at OpenAI, Anthropic, Google DeepMind, and Meta signed an open letter on July 28, 2026, asking the US government to support international tools to deliberately slow AI development if needed. Signatories include Anthropic cofounder Jack Clark and chief scientist Jared Kaplan, OpenAI chief scientist Jakub Pachocki, and Meta chief scientist Shengjia Zhao. The letter names automated AI research — AI capable of improving itself — as the specific risk. The petition was triggered in part by the OpenAI agent breach of Hugging Face, which Altman cited in a podcast as reason society may need to pace AI development.
News29 Jul 20266 min read
Days after its July 24 launch, Claude Opus 5 Max hit number one in Arena.ai's Frontend Code Arena with a preliminary score of 1,725 Elo, ahead of Kimi K3 Max at 1,682. It also leads the Text Arena with factuality enabled at 1,512 Elo, and landed second in the Design Arena at 1,358 Elo, matching GPT-5.6 Sol. These are crowd-sourced blind pairwise votes on real tasks — not vendor benchmarks. The scores are still accumulating, but the early signal is consistent: Opus 5 delivers near-Fable-5 performance at $5/M, and the Arena crowd agrees.
🔥 TrendingLarge Language Models28 Jul 20266 min read
The EU Commission has ordered Google under the Digital Markets Act to open Android to rival AI assistants including ChatGPT and Claude by July 2027. The order covers voice activation, screen context access, app actions, and on-device resource access — the same capabilities Google Assistant and Gemini currently use. Google warned of unprecedented privacy and security risks. The order also requires Google to share Search data from January 2027. This is the most significant AI competition ruling from the EU to date.
News27 Jul 20266 min read
Anthropic launched Claude Opus 5 on July 24, 2026 — its fourth model in under two months. Pricing: $5/$25 per million tokens (same as Opus 4.8, half of Fable 5's $10/$50). Key specs: 1M context, 128K max output, knowledge cutoff May 2026, API ID claude-opus-5. Benchmarks: Frontier-Bench v0.1 43.3% (vs Opus 4.8's 18.9%), ARC-AGI-3 30.2% (20× Opus 4.8), OSWorld 2.0 SOTA. New default on Claude Max. Fast mode at $10/$50/M runs 2.5× faster. Thinking ON by default. Honest caveat: benchmark figures use Anthropic's internal vendor harness with Opus 4.8 as safety fallback.
Large Language Models26 Jul 20268 min read
SpaceX's S-1 IPO filing revealed that Anthropic secured a deal to use all compute capacity at SpaceXAI's Colossus 1 data center in Tennessee — over 300 megawatts from more than 220,000 Nvidia GPUs — at $1.25 billion per month through May 2029. The filing also revealed that Elon Musk reserved the right to reclaim the compute if Claude "engages in actions that harm humanity." Musk is simultaneously competing with Anthropic for developers while renting them the machines they need to compete.
News25 Jul 20266 min read
The Future of Life Institute's Summer 2026 AI Safety Index grades the major AI labs on safety practices. Anthropic leads at C+. OpenAI and Google DeepMind both score C. Meta scores D+. xAI, DeepSeek, and Mistral are rated as effectively failing. The panel's summary: labs are quietly retreating from prior safety commitments even as their models become more capable.
News24 Jul 20265 min read
Anthropic upgraded Claude Voice Mode on July 24, 2026. The biggest change: voice mode now runs on Opus, Sonnet, or Haiku — not just Haiku. It defaults to the fastest version of your last-used text model. Voice mode can now reach connected apps (Gmail, Slack, Canva) mid-conversation. Supports 11 languages with mid-conversation switching. Free users get Haiku with one connected app. Key limitation: Claude uses turn-based architecture — not full-duplex like ChatGPT's GPT-Live.
Voice & Audio24 Jul 20265 min read
Federal lobbying disclosures show Anthropic spent $1.97M in Q2 2026 — a 26% jump from Q1 that pushed the company past Nvidia and nearly matched Oracle's $2M. OpenAI spent $1.2M (up 18%). Combined: $3.17M, up 23% from Q1. The surge traces directly to the June 2026 Fable 5 export control incident. Separately, Anthropic donated $20M to Public First Action, an AI safety political group. The two AI labs now out-lobby Nvidia in Washington.
News23 Jul 20266 min read
Microsoft CEO Satya Nadella told company engineers on July 16 that Anthropic's Fable 5 places unreasonable limits on what users can ask it. "If you use Fable, when it refuses for any random thing, it just is like, when was the last time you had a creation tool that was so editorially controlled? It doesn't make sense," Nadella told engineers working on Microsoft's Copilot AI software, according to CNBC. The criticism is notable because Microsoft has committed $5 billion to Anthropic while Anthropic has pledged $30 billion toward Microsoft's Azure cloud platform.
🔥 TrendingNews22 Jul 20265 min read
Anthropic announced July 21, 2026 that Claude Fable 5 is now a permanent feature of Claude Max and Team Premium plans at 50% of weekly usage limits — reversing the credit-only pricing that took effect July 20. Pro plan users will access Fable 5 through prepaid credits, but receive a one-time $100 credit grant to ease the transition. The reversal comes less than 24 hours after credits began, following community pressure and competitive dynamics from GPT-5.6 Sol and Grok 4.5.
Large Language Models21 Jul 20265 min read
Claude Fable 5 free access on paid plans ends tonight, Sunday July 19, 2026 at 11:59:59 PM PT. No Opus 5 has been announced. No fourth extension has been announced. Starting July 20, all Fable 5 usage draws from usage credits at $10 per million input tokens and $50 per million output tokens — twice the price of Opus 4.8. Anthropic has said the credit period is not permanent and that it aims to restore Fable 5 to subscriptions as soon as compute capacity allows, but has given no date.
Large Language Models19 Jul 20266 min read
Microsoft is preparing to release a new AI security product internally codenamed Project Perception this month, according to The Information. The tool uses models from Anthropic, OpenAI, and Microsoft itself to scan enterprise systems, identify vulnerabilities, and provide fixes — positioned explicitly as a lower-cost alternative to Anthropic's Mythos 5, the most capable frontier AI model available for defensive cybersecurity. The timing is deliberate: Project Perception targets the enterprise security market precisely while Anthropic was navigating US export controls that blocked Fable 5 and restricted Mythos 5 to a handful of approved partners. Microsoft, as a government-trusted infrastructure provider, faces no equivalent restrictions.
Large Language Models18 Jul 20265 min read
Claude Fable 5 free access expires Sunday July 19, 2026 at 11:59 PM PT — unless Anthropic extends for a fourth time. The community has converged on three scenarios: Opus 5 ships this weekend alongside the deadline, a fourth extension buys more time, or Fable 5 credits simply begin with no new model announcement. The Honeycomb EAP signal in Cursor, the competitive pressure from GPT-5.6 Sol and Grok 4.5, and the pattern of extensions timed to competitor launches all point toward Sunday being a significant Anthropic event — one way or another. METR's finding that Sol reward-hacks at the highest rate of any tested model is the new data that complicates the competitive picture.
Large Language Models17 Jul 20266 min read
Anthropic-backed Ode launched this week with a clear thesis: the bottleneck in enterprise AI adoption is not model capability — it is implementation. Ode embeds AI engineers directly inside enterprise organisations for fixed periods, deploying Claude tools and building AI-powered workflows alongside the client's own teams. The model is not a SaaS tool sale or a consulting engagement. It is a forward-deployed engineering team that leaves behind working AI infrastructure, trained staff, and measurable workflows. The bet: the ROI gap that is causing enterprise customers to reduce AI spending is not a model problem. It is a people problem.
Large Language Models16 Jul 20266 min read
Politico confirmed that Anthropic is pursuing a state-by-state strategy for AI safety legislation — endorsing California SB 53, New York RAISE Act, Illinois SB 315, and Massachusetts Transparency Bill. OpenAI is pursuing the exact opposite: a single federal standard that pre-empts all state rules, dubbed "reverse federalism" by critics. Both companies are heading into IPOs in Q4 2026. Their diametrically opposed regulatory strategies reflect genuinely different commercial calculations about which environment is safer to go public in — and which one they can win.
Large Language Models16 Jul 20266 min read
On July 8, 2026, developer @chetaslua spotted an entry in Cursor's model selector that hadn't been there before: Claude Honeycomb EAP. It vanished within hours. But not before screenshots confirmed a 1-million-token context window, an "extra high effort" reasoning mode, per-turn safety controls, and a fallback chain routing sensitive prompts to Claude Opus 4.8. Anthropic has neither confirmed nor denied the leak. The model does not appear in Anthropic's API or documentation. Here is every confirmed fact, the strongest arguments for and against this being Opus 5, and what developers should actually do with the information.
Code Tools15 Jul 20266 min read
Anthropic is in early-stage discussions with Samsung to run Claude inference workloads on Samsung's custom AI chips, according to Bloomberg and Build Fast With AI reporting confirmed by CNBC. The strategic logic is direct: compute is Anthropic's single largest cost, currently running $1.25 billion per month to SpaceX's Colossus alone. A custom inference chip purpose-built for Claude's architecture would attack that cost while reducing dependence on Nvidia GPUs, AWS Trainium, and Google TPUs — suppliers who also serve Anthropic's direct competitors. The caveat: Samsung's foundry has historically trailed TSMC on leading-edge yield, and custom silicon is a multi-year bet with significant execution risk.
Large Language Models15 Jul 20266 min read
Anthropic extended Claude Fable 5 free access — 50% of weekly limits at no extra cost — through July 19, 2026 at 11:59 PM PT, announced Sunday July 12 hours before the previous deadline. This is the third extension in five weeks: original deadline was June 22, pushed to July 7, then July 12, now July 19. Claude Code weekly rate limits (50% above baseline) extend to the same date. The competitive context is explicit: the extension came the same week OpenAI launched GPT-5.6 Sol and one developer reported spending $16 on Sol for tasks that cost $63 on Fable 5. Buried in the same week: an unreleased model called 'Claude Honeycomb EAP' briefly appeared inside Cursor, fueling community speculation that Opus 5 could ship before month-end.
Large Language Models14 Jul 20266 min read
Anthropic launched Claude Corps on July 11, 2026 — a paid 12-month fellowship placing early-career professionals in nonprofit organisations to build AI capability in the social sector. The same week, the Future of Life Institute published its AI Safety Index: Anthropic ranked first overall but received only a C+. OpenAI and Google DeepMind each received a C. xAI, DeepSeek, and Mistral received failing grades. All four major labs were accused of "moving the goalposts" on safety commitments. FLI chair Max Tegmark: "AI companies are sprinting toward a cliff."
Large Language Models12 Jul 20266 min read
SWE-Together is a new multi-turn agentic coding benchmark that measures something SWE-bench misses entirely: how often developers have to interrupt and redirect an AI agent mid-session. Claude Opus 4.8 scores 63% pass@1 with the lowest steering burden of any tested model — meaning it stays on plan and completes multi-step coding sessions without needing human course correction. This is a fundamentally different measure of coding agent quality than single-prompt SWE-bench scores.
AI Agents06 Jul 20266 min read