BIG TECHAWAYS

WHAT MATTERS TODAY

​Anthropic makes Fable 5.1 cheaper and more private​

Anthropic has moved Fable 5.1 into general availability and made Mythos 5.1 available through trusted-access programs. The company says typical token-billed workloads are about 25% cheaper, while highly agentic workloads can be up to about 45% cheaper.

The change is not only about price. Enterprise Frontier Safeguards let customers keep data in cloud infrastructure they control while Anthropic maintains misuse-monitoring mechanisms. Frontier models are increasingly being sold as a package of capability, economics, privacy, and safety controls, rather than as a benchmark score alone.

The benchmark and customer-performance results in the announcement come from Anthropic. The company also notes that safeguard interventions can affect some scores.

​OpenAI rates Astra Critical for cyber capability​

OpenAI says Astra can find previously unknown vulnerabilities and develop exploits across well-protected systems without step-by-step human guidance. The model will be released with stronger safeguards, including restricted access to its most advanced cyber capabilities.

For models that can perform complex cyber work autonomously, access policy, monitoring, and capability limits become part of the product. They determine where the model can be used and how much of its capability is available. This is an internal evaluation published by OpenAI, not evidence that Astra attacked real systems.

​AI data-center spending could reach $31.6T​

Bloomberg, citing a PwC report, says the central scenario puts global data-center spending at $31.6 trillion through 2050. If AI adoption accelerates, the figure could reach $50 trillion. The US is modeled to capture about $15.1 trillion.

Much of that investment will not go only toward land, buildings, and power systems. GPUs, servers, storage, and networking equipment will require regular replacement, making this a recurring refresh cycle rather than a one-time construction boom. PwC projects annual spending to rise from about $800 billion this year to $1.1 trillion in 2030 and $1.8 trillion in 2050. These are scenarios modeled by PwC and Oxford Economics, not guaranteed spending levels.

PRESENTED BY VIKTOR

Six people doing the work. Your headcount is one.

Your finance close runs in #finance. Stripe and QuickBooks reconciled, runway updated, posted Sunday night without you asking. Engineering review lands in #eng.

​Viktor pulled the open PRs, left comments on auth-refactor, flagged a dependency blocking api-pagination.

Campaign brief lands in #growth: Meta CPA up 18%, recommendation to pause broad match, a draft landing page already deployed for the variant test. You hired him on day zero. He lives in Slack and Microsoft Teams alongside your contractors and investors, connects to 3,000+ tools, pushes back when you ship something dumb.

Viktor is now an integral team member, and after weeks of use we still feel we haven't uncovered the full potential." Patrick, Director, Yarra Web.

​Get Started for Free​

BIG THINK

​The AI boom is also a bet on reflexive demand​

Giovanni Cattani divides labor demand along two axes: bounded versus unbounded work, and short-horizon versus long-horizon work. Tax preparation is a bounded example, while AI research, software development, space exploration, and trading are unbounded activities where there is always more work to do.

From that framework, Cattani argues that much of today’s frontier-token demand is concentrated in AI R&D, software engineering, and trading. He uses METR to illustrate the task horizon a model can handle with a high probability of success, contrasting roughly 30 minutes for o3 with about three hours for Mythos. He estimates that the three categories could represent roughly 20%, 15%, and 15% of frontier-lab inference revenue.

These are Cattani’s framework and estimates, not audited market shares. They raise a useful question: does AI demand come from stable end users, or is part of it amplified by a loop between token spend, revenue, valuations, and access to capital?

SHIFT SIGNALS

BEHIND THE HEADLINES

• ​Microsoft turns responsible AI into an operating report: Microsoft published its 2026 Responsible AI Transparency Report, covering governance, tools, customer deployment, reliability, and security. As models move deeper into enterprise products, the ability to explain how they are governed becomes part of the trust infrastructure.
• ​Gemini goes deeper into everyday Android utility: Google is adding Guided Vision, remembered items in Find Hub, and Motion Assist to Android. Small, repeated features embedded in an operating system can distribute AI more widely than a standalone chatbot.
• ​NVIDIA and CrowdStrike put defensive AI into an offense-defense loop: SafeMind combines cyber models, custom harnesses, and continuous offense-defense testing, according to NVIDIA and CrowdStrike. Defensive AI is being designed as a system that constantly tests itself, not just as a model that detects threats.
• ​Meta brings Muse Voice Transcribe into the product layer: Mark Zuckerberg says Muse Voice Transcribe supports speaker diarization, endpointing, adaptive delay, more than 70 languages, hour-long sessions with more than 20 speakers, and biasing for names or specialist terms. The model is already used in the Meta desktop app, Muse Code, and the Meta model API with a zero-data-retention tier. Performance claims come from Meta.
• ​MI355X performance on agentic workloads is improving through software: SemiAnalysis says AMD’s team improved MI355X performance on realistic workloads using Qwen3.5 397B and MiniMax M3 within a few weeks, and expects the gap with B300 to keep narrowing. This is SemiAnalysis’s assessment, not an independently reproducible benchmark from the thread.
• ​Qwen is updating models faster than version numbers: Qwen3.8-Max-0902 is described as having 2.4 trillion parameters, a 1 million token context window, and post-training for Coding and Cowork, based on the Qwen announcement quoted in the post. TechNode also reports the update. The pace of these post-training snapshots shows how model competition is increasingly happening through frequent, incremental changes.

PRESENTED BY INSURIFY

Are you paying too much for your car insurance?

Insurance rates have climbed, and if you haven't compared recently, there's a good chance you're overpaying. Insurify lets you see real quotes from top carriers in minutes, so you know exactly where you stand. Just enter your car make and ZIP code to instantly pull up rates side by side. No phone calls, no agents, no fees — just a fast, clear look at what's available in your area. Drivers are finding rates starting as low as $39/month. The only way to know is to look.

​See Rates on Insurify​

Average potential savings based on initial quotes received by 151,494 customers seeking insurance through Insurify. Actual savings may vary depending on state of residence, individual circumstances, coverage selections, and insurance provider. Savings and lowest rates do not reflect typical results.