Anthropic cleared $11.5B in Q2 revenue and posted positive adjusted operating income, while DeepSeek raised some V4-Pro API prices by up to 1,100% effective today. Subsidized inference is ending.
Anthropic posted preliminary Q2 2026 revenue above $11.5 billion and its first positive adjusted operating income, according to investor documents reported by Bloomberg. The milestone lands as IPO chatter circulates and as rivals abandon loss-leader pricing.
Writer rebuilt the software surrounding its agents rather than swapping models, cutting cost per task roughly 41%. Nvidia's NeMo Switchyard work points the same direction: orchestration, not model choice, is the lever.
In his August 15 Independence Day address, Prime Minister Narendra Modi announced an AI skilling program targeting one crore Indian young people within a year. Talent supply is now state policy.
Koda publishes a fact-checked AI intelligence briefing every morning: site, email, podcast, and video. Founding sponsors lock today's rates for 6 months as the audience compounds.
View the media kit →Anthropic posted preliminary Q2 2026 revenue above $11.5 billion and its first positive adjusted operating income, according to investor documents reported by Bloomberg. The milestone lands as reports circulate about an eventual Anthropic IPO that would test public-market appetite for the AI buildout. Profitability at that revenue scale undercuts the assumption...
Continue reading arrow_outwardAnthropic posted more than $11.5 billion in preliminary Q2 2026 revenue and its first positive adjusted operating income, according to investor documents reported by Bloomberg, with IPO chatter now circulating around the company.
Lebanon's state news agency said an Israeli airstrike hit a house on the outskirts of Ansar early Saturday, killing seven people and wounding three, with the day's total toll from Israeli attacks reaching 11 dead.
Mood reads Fear even as Anthropic demonstrates frontier labs can stop burning cash, a disconnect that says investors are pricing the buildout, not the earnings.
Apple training a China-only model with Alibaba and Modi pledging AI skilling for 10 million youth are the same story: national borders are becoming model architecture.
DeepSeek raised some V4-Pro API prices by up to 1,100% effective today, betting agentic coding gains justify premium rates over volume. Anthropic's profitable $11.5 billion quarter proves the other half of the thesis: frontier labs do not have to burn cash indefinitely. Together they undercut the assumption that inference costs only fall.
Apple abandoned wholesale licensing of Chinese partners' models and trained its own for the Chinese market with Alibaba's support, a decision driven by regulation rather than performance. India, meanwhile, is Claude's second-largest market and Modi is promising to skill 10 million young people within a year. Compute and compliance are now drawn along the same map.
Honor's August 12 launch in Guangzhou paired a 4-degree-of-freedom mechanical gimbal with AI motion control and ARRI Image Science tuning, calling it the first robotic smartphone. The bet is that hardware physically responding to model output is the next differentiator once software features commoditize. Watch whether rivals follow or let Honor absorb the manufacturing risk alone.
Deep Dive
7.6/ 10
Most chat agents are hostages to a browser session; this one treats the conversation as a background job with retries, approvals, and logs.
This is for TypeScript teams who already think in terms of queues and workers and want their agent runs to survive refreshes, redeploys, and crashes. If your agent work looks like migrations, batch cleanup, or multi-step scrapes rather than snappy Q and A, it is worth adopting; if you want a no-code agent builder or a drag-and-drop canvas, look elsewhere.
“The launch framing is blunt about the pitch: AI chat that keeps running after you close the tab, streaming through refreshes and crashes.”Product Hunt
Kilo Code writes, reviews, fixes, and explains code, and because the project is open source you can inspect how it handles your repository instead of guessing. Point it at a small module first and use the review and explain paths before you let it write anything you plan to ship. The open license also matters if your employer blocks closed assistants that phone home.
Try it arrow_outwardBallet builds agentic workflows designed to produce the same outcome on every run, trading flexibility for consistency. That is the correct tradeoff for invoice handling, onboarding checklists, and reporting, where a creative answer is a defect. Test it by running the same input ten times and comparing outputs; if the variance is zero, it is doing the one thing it claims.
Try it arrow_outwardLaraCopilot pitches agentic generation of production-ready applications rather than autocomplete inside your editor. Use it for prototypes and internal tools where a rewrite is cheap, and read the generated auth, database queries, and dependency list yourself before anything touches real users. The gap between a demo app and a production app is almost always in those three places.
Try it arrow_outwardSidekick is surfacing this week as an agent interface for Mac users aimed at autonomous multi-step tasks on your own machine, which means file access and app control rather than cloud-only chat. That is useful for repetitive local work like renaming, sorting, and moving data between desktop apps. It appeared in product digest coverage rather than official release notes, so verify permissions scope on the product page before you grant it your file system.
Try it arrow_outwardRe-cost every agent loop today. Pull your last 30 days of API spend and reprice it against DeepSeek's new V4-Pro rates effective August 16. Anything that only worked at subsidized pricing needs a new plan this week.
Rewrite the software, not the model. Copy Writer's approach: before swapping models, cut redundant tool calls, cache aggressively, and trim context. Writer got roughly 41% off cost per task without changing the model.
Test an auditable and durable agent pair. Run Kilo Code, the open-source coding assistant you can read before you trust, alongside Trigger.dev's Chat Agent, which keeps working after you close the tab. Compare them on one real backlog ticket.
Anthropic proved frontier AI can turn a profit at $11.5 billion a quarter, and DeepSeek immediately stopped pretending inference is free. For builders, that means the arbitrage of cheap tokens is closing and efficiency work moves from nice-to-have to survival. Writer's 41% cut shows the gains are still there, but they live in your orchestration layer rather than in a vendor's price list. Meanwhile Apple's China-only model and Modi's ten million skilling pledge confirm the second half of the story: capability is going sovereign, and so is the talent that runs it.
AI analysis, world news, markets, and tools. One briefing, delivered free.