Z.ai says GLM-5.3 weights land inside two weeks and Alibaba has already opened Qwen3.8-2.4T-A95B. Senator Jim Banks answered on August 14 with a letter asking the administration for incentives.
Z.ai told Bloomberg that GLM-5.3, built on the same roughly 700-billion-parameter base as GLM-5.2, ships weights inside two weeks. Alibaba already opened Qwen3.8-27B and the 2.4-trillion-parameter Qwen3.8-2.4T-A95B mixture of experts.
OpenAI's GPT-5.6 builder playbook is mostly about which tier to call and when to trade reasoning depth for latency. Gemini 3.7 Flash scores 56 on the Artificial Analysis index at 340 tokens per second, second fastest in the set.
On Aug 14 the White House said it had identified India and 40 other countries in a new enforcement push against suspected tariff evasion involving Chinese goods routed through third markets. It lands the same week India became Claude's second-largest market.
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 →Z.ai (Zhipu) told Bloomberg it is preparing GLM-5.3, built on the same roughly 700-billion-parameter base as GLM-5.2 but tuned for coding, with internal benchmarks placing it close to and sometimes ahead of Anthropic's Fable 5 on coding evaluations. The company said it plans to publish the model's weights within...
Continue reading arrow_outwardZ.ai says GLM-5.3, built on the same roughly 700-billion-parameter base as GLM-5.2 and tuned for coding, will have downloadable weights within two weeks, with internal benchmarks near or ahead of Anthropic's Fable 5.
Trump told a New York rally on Aug 14 that he will "pretty soon" declare the Strait of Hormuz US territory, as Iran denies access to much civilian shipping and the US Navy blockades Iranian ports.
The S&P 500 and Nasdaq closed at reported record highs on Aug 14 after cooler US producer prices, yet mood stays Fear with WTI elevated on Hormuz risk and gold selling off sharply.
Senator Jim Banks wants incentives for US open weights while the White House names India and 40 others in a transshipment probe: Washington is trying to slow Chinese hardware routes and speed American software giveaways at the same time.
Z.ai plans to publish GLM-5.3 weights inside two weeks, and Alibaba has already released Qwen3.8-27B plus the 2.4-trillion-parameter Qwen3.8-2.4T-A95B mixture of experts with roughly 95 billion active parameters. The competitive question is no longer whether a frontier-class model exists but how quickly its weights hit a download client. American labs are shipping open entries too, including Meta's Muse Glimmer and Nvidia's Nemotron line, but the release cadence is coming from Hangzhou and Beijing.
Banks released his letter on Friday, August 14, urging the Trump administration to build incentives for US open-weight development, an argument framed around ceded ground rather than any concrete program. Letters move slower than model releases: GLM-5.3 weights are due within two weeks. The gap between legislative pressure and shipping schedules is the story to watch.
OpenAI's GPT-5.6 builder guide is explicitly about which tier to call and when to trade reasoning depth for latency, following the Ultrafast preview of GPT-5.6 Sol at up to 14 times prior speed. Google's Gemini 3.7 Flash makes the same case from the other direction: 56 on the Artificial Analysis Intelligence Index, ninth overall, at 340 tokens per second and two to three times cheaper per API call than Claude Sonnet. Developers are being handed dials, not just bigger models.
Deep Dive
6.8/ 10
It is a model buffet with a credit meter, and the smartest way to use it is to generate ads you fully intend to delete.
This is for marketers and solo founders who need to see three hook variations before committing budget to a shoot or an editor. It is a competent aggregator with a clean single workspace and a credible model lineup, but it is a concepting tool, not a finishing tool, so treat every output as a storyboard that happens to move. Worth the Pro seat if you are testing creative weekly; skip it if you need broadcast-ready assets out the other end.
“Simple to use, the video quality holds up, and having several AI video models under one roof is the appealing part.”Third-party AI tool directory listing, surfaced in research
Lovable takes a prompt and ships a running full-stack web app, with GitHub and Supabase wiring handled for you, then lets you edit and deploy from the same surface. The realistic use is MVPs and internal tools where the database schema is simple and the exit cost of a rewrite is low. Its Product Hunt listing shows enterprise-only pricing, so treat it as a team purchase rather than a weekend experiment.
Try it arrow_outwardPlay lets you design and prototype mobile products directly on an iPhone using real iOS materials, so you test motion and touch targets on the device instead of guessing from a desktop artboard. It is free, which makes it a cheap way to sanity check a screen before you commit it to a design system. Best used for the last-mile feel check, not for maintaining a shared component library.
Try it arrow_outwardUgic is a Figma plugin that reads your existing components and design system, then converts text such as a PRD into structured, multi-language layouts. That constraint is the point: output lands inside your system instead of producing off-brand mockups you have to rebuild. It is a paid plugin, so scope one real feature spec at it before rolling it out to a design team.
Try it arrow_outwardRupt monitors for shared logins with identity verification and fraud signals, then routes those users toward paying for their own seat rather than simply locking them out. For a small SaaS, the leverage is in the conversion flow, not the detection, so tune the messaging before you enable enforcement. The Scale plan offers 14 days free with no credit card, which is enough time to measure how much sharing you actually have.
Try it arrow_outwardWrite a routing table for GPT-5.6. Use OpenAI's new builder playbook to map three job types in your product to a specific tier, then log latency and cost per call so the trade between reasoning depth and speed is measured rather than guessed.
Pin an open-weight fallback. Stand up Qwen3.8-27B locally this weekend and queue capacity for GLM-5.3 weights, which Z.ai says land inside two weeks. Cheap open tiers now run two to three times cheaper than Claude Sonnet 5 or GPT-5.6 Terra.
Prototype the interface, not the deck. Try Lovable for a full-stack pass, Play for native iOS layout on the device itself, and Ugic to generate Figma screens from your own component library. The Scale plan offers 14 days free with no credit card.
Two Chinese labs are publishing near-frontier weights inside a fortnight while the US policy response is a letter dated August 14. For builders that asymmetry is an opportunity, not a headline: open Qwen and GLM tiers undercut Claude Sonnet 5 and GPT-5.6 Terra by two to three times on price, and Gemini 3.7 Flash shows that latency now competes with raw index score. The skill that compounds this quarter is routing, which tier answers which request, measured in tokens per second and dollars. Markets drifted lower and crypto sentiment sat at 34 in Fear, so nobody is paying a premium for your infrastructure story right now. Ship the cheap version that works.
AI analysis, world news, markets, and tools. One briefing, delivered free.