Alibaba's Qwen team says Qwen3.8-Max weights go public next week, a 2.4-trillion-parameter model with a 1-million-token context window. The paid API stays at $2 per million input tokens. Frontier scale just stopped being a moat.
Qwen3.8-Max ships its weights publicly next week: 2.4 trillion parameters, 95 billion activated, a 1-million-token context window. Bloomberg reports it matches or exceeds Anthropic's flagship Fable 5 on top global benchmarks, and the paid API still lists at $2 per million input tokens and $6 per million output.
Samsung announced next-generation AI memory technology on August 4, extending the bandwidth race. Memory capacity per accelerator, not compute, decides how large a model you can actually train or serve.
Qatar said on Aug 4 that mediators are making progress toward ending the US-Iran war, sending Brent down more than 5% on top of the previous day's losses. Falling energy costs plus strong AI-linked earnings pushed the Dow and S&P 500 to record closes.
A weekly operator deep dive, the full Dojo Pro prompt packs, and the complete prompt database. Founding members lock the launch price forever.
Join the founding list →Alibaba's Qwen team says the weights for Qwen3.8-Max, the 2.4-trillion-parameter model with 95 billion activated parameters and a 1-million-token context window, will be made publicly available next week, alongside an open-source release of the smaller Qwen3.8-27B. That would put a model Bloomberg reports matches...
Continue reading arrow_outwardAlibaba's Qwen team says weights for Qwen3.8-Max, a 2.4-trillion-parameter model with 95 billion activated parameters and a 1-million-token context window, go public next week alongside an open-source Qwen3.8-27B.
Qatar said on Aug 4 that mediators are progressing toward ending the US-Iran war and Treasury Secretary Scott Bessent hinted a Hormuz reopening deal could be near, though Tehran denies talks are under way and the strait remains effectively shut.
The Dow and S&P 500 closed at record highs Tuesday as AI-linked earnings met a 5% drop in Brent crude, yet the mood reading stays at Fear because the rally rests on a peace deal Tehran has not confirmed.
Qatar said on Aug 4 that mediators are making progress toward ending the US-Iran war, sending Brent crude down more than 5% on top of the previous day's losses, and US Treasury Secretary Scott Bessent suggested a deal to reopen the Strait of Hormuz could be near.
Publishing Qwen3.8-Max weights would put a model Bloomberg reports matches or exceeds Anthropic's flagship into unrestricted circulation, with the smaller 27B variant open-sourced for anyone to run locally. The strategic effect is to make the frontier a commodity rather than a licensed service. Every US lab pricing per token now competes against a free download.
OpenAI's publication of third-party cyber evaluations lands while the White House favors voluntary safety testing over mandates, which makes external red-team reports the primary evidence regulators will ever see. That gives labs control over both the testing scope and the disclosure. It is a weaker check precisely when open weights remove the ability to revoke access after the fact.
Samsung's next-generation AI memory announcement targets the real bottleneck, since memory capacity per accelerator, not logic, now limits how large a model any given fleet can train or serve. The Sichuan floods that forced more than 900,000 evacuations expose the other constraint: hydropower and the data center capacity depending on it. Compute scarcity is increasingly a supply-chain and weather story.
The update is explicitly about cutting input-token usage and making large agent deployments cheaper to run, which matters more than latency once you have dozens of agents in production. Before upgrading, log the input tokens per task for your three most-used agents so you have a baseline to compare against. Context bloat, not model choice, is usually where agentic app costs escape.
Try it arrow_outwardSumly.AI turns podcasts, talks, and recorded calls into condensed summaries, and it is currently free and ad-free. The practical use is triage: run the 90-minute conference session or earnings call through it first, then decide whether the full recording earns your time. Keep the summary as a searchable note and cite timestamps back to the original before you quote anything.
Try it arrow_outwardFinalle.ai aggregates new-media and financial data streams and layers generative summarization on top, aimed at teams that need faster reads on markets and companies. Test it against a week you already understand well and score how often its signals matched what actually mattered. Any number it surfaces should be checked against filings or primary data before it enters a document anyone acts on.
Try it arrow_outwardCarta's 3 August analysis of the startup exit environment concludes that the split between companies with promising exits and those without usually comes down to AI-driven growth. Read it as a diagnostic on your own metrics: whether AI is producing measurable revenue or retention movement, not whether it appears in your deck. If you cannot point to the line it moved, buyers will not either.
Try it arrow_outwardWith EU AI Act enforcement widening from 2 August, the volume of consultant explainers has outpaced the accuracy of them. The Commission's own policy page is the canonical map of obligations, timelines, and the linked funding and research programs, and it is updated as guidance lands. Bookmark it and check the primary text before you change a product decision on the basis of a blog post.
Try it arrow_outwardPrice your stack against $2 in. Benchmark your current provider's cost per task against Qwen3.8-Max at $2 per million input and $6 per million output tokens, then decide whether next week's open weights change your hosting math or just your negotiating position.
Cut your input-token bill first. Run LangChain Deep Agents against one long-context workflow and measure token spend before and after. A 1-million-token window is only cheap if you are not stuffing it blindly.
Build a real evaluation harness. OpenAI published third-party cyber evaluations while Washington favors voluntary testing over mandates, so external red-team reporting is the only public signal. Write down the three failure modes you will test on any model you adopt, including open weights you host yourself.
A 2.4-trillion-parameter model going public next week does not hand you a business, it hands you a commodity input. The scarce things are memory bandwidth, which is why Samsung's August 4 announcement matters more than it reads, and credible evaluation, which is why OpenAI's third-party cyber disclosures land in a voluntary-testing vacuum. Markets already priced the optimism with the Dow and S&P 500 at record closes on cheaper crude and AI earnings. Your job today is to know exactly which layer of that stack you own.
AI analysis, world news, markets, and tools. One briefing, delivered free.