Top Story
Lead
Anthropic's Mythos Leaks: The Model Race Has a New Leader
AI Models · Sunday, March 29 · JPM Desk
Internal benchmarks for Anthropic's unreleased Mythos/Capybara model surfaced this week. The numbers are rewriting the competitive map.
The leaked data positions Mythos as Anthropic's most capable model to date, outperforming current flagship benchmarks across reasoning, code, and long-context tasks. The timing is deliberate: Anthropic is in IPO positioning mode, and a leaked capability ceiling is as good as a press release.
For marketers, this matters beyond the model specs. The Mythos leak lands alongside Claude Computer Use going live on macOS and Anthropic eliminating the long-context pricing surcharge, making the 1M token context window standard pricing. Anthropic is stacking capability, automation, and cost signals simultaneously.
Read Full Story · Fortune
1M
Token context window now at standard pricing, long-context surcharge eliminated · Anthropic
That pricing move is underrated. The previous surcharge made extended-context use cases a premium decision. Removing it opens the door for agencies and operators to build against large documents, full campaign histories, and deep research tasks without the cost math working against them.
"Capability leak. Automation launch. Price drop. Anthropic is making three moves at once."
Also This Week
Claude Computer Use Launches on macOS
Anthropic's automation play is live. Claude can now operate desktop applications, navigate interfaces, and execute multi-step tasks on Mac. The practical ceiling for no-code automation just moved significantly higher for anyone already in the Claude ecosystem.
Read More · VentureBeat
Anthropic Removes Long-Context Pricing Surcharge
1M token context is now standard pricing with no premium. Developer-friendly positioning ahead of what looks like a big capability push. The combination of Mythos specs and this pricing move signals Anthropic is competing hard on both performance and accessibility.
Read More · Byteiota
Model Race
Competitor Watch
OpenAI's Spud Completes Pre-Training as Sora Is Officially Shut Down
AI Models · OpenAI
OpenAI confirmed pre-training is complete on Spud, its next frontier model. Sam Altman's framing: this model won't just assist work, it will accelerate the economy. The positioning is shifting from tool to infrastructure.
Meanwhile, Sora is officially shut down. The open question now is whether its video capabilities get quietly folded into ChatGPT or another product down the line. If you built workflows around Sora, it's time to find a replacement.
Read Full Story · The Decoder
"Two frontier models in the same week. The gap between labs and everyone else keeps widening."
Anthropic Updates
Claude Computer Use Launches on macOS
Anthropic's automation play is live. Claude can now operate desktop applications, navigate interfaces, and execute multi-step tasks on Mac. The practical ceiling for no-code automation just moved significantly higher for anyone already in the Claude ecosystem.
Read More · 9to5Mac
Long-Context Surcharge Eliminated + Faster Rate Limits at Peak
1M token context is now standard pricing with no premium. Anthropic also announced faster rate limits during peak hours, a direct response to developer complaints about throughput. Both moves signal Anthropic is competing hard on accessibility, not just capability.
Read More · The Register
Commerce
Live Now
Shopify Agentic Storefronts Go Live Inside ChatGPT
5.6 million Shopify merchants can now be discovered, browsed, and purchased through ChatGPT conversations. Commerce has entered the AI chat interface and product discovery is being rewired in real time.
Read More · Shopify
Platforms
Expanding
Google & YouTube: Gemini Now Selects Your Influencers
Gemini is being embedded into YouTube's creator marketplace to automate influencer matching, analyzing audience fit, brand alignment, and performance signals without a human in the loop. Another manual layer of campaign management that just went autonomous.
Read More · AdWeek