The Manifest
Anthropic and OpenAI both raced to cut prices on new flagship models while container schedule reliability and Rhine barge levels worsened for planners, and fresh capital kept flowing into AI training data, freight tech and port infrastructure deals.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01Anthropic releases Opus 5.5 with lower prices and Fable-level performanceTechCrunch
- 02OpenAI launches GPT-6 Sol and Luna, boasting lower cost and fewer mistakesTechCrunch
- 03New report: Just a third of container shipping on-timeFreightWaves
- 04Snorkel AI triples valuation to $3.5B as demand for AI training data boomsTechCrunch
- 05Pentagon says overreliance on AI contributed to missile strike on Iran schoolbloomberg.com
Models & Releases
5 storiesAnthropic releases Opus 5.5 with lower prices and Fable-level performance
Anthropic cut Opus pricing about 40 percent while claiming to beat GPT-6 Astra on most benchmarks, another sign the frontier race is shifting to cost per token rather than raw capability. Anyone renewing an AI vendor contract should recheck unit economics before signing.
OpenAI launches GPT-6 Sol and Luna, boasting lower cost and fewer mistakes
The new models halve token prices but independent testing shows little actual intelligence gain, so this reads as a pricing response to Anthropic rather than a capability leap. Budget for it accordingly.
Xiaomi's affordable flagship AI leads the open models, and Anthropic says Claude helped get it there
MiMo-V2.6-Pro tops open model rankings at a fraction of typical training cost, but Anthropic is accusing Xiaomi of scraping Claude outputs to get there. Worth flagging for teams weighing open-weight models against data provenance risk.
Qualcomm launches two new smartphone chips with emphasis on AI
Running a 30B mixture-of-experts model on-device pushes more inference off cloud infrastructure and onto handsets, which matters for anyone budgeting cloud AI spend tied to edge deployments.
OpenAI hires Patreon co-founder Sam Yam to lead a new Creator Product division
A new product push aimed at creators signals ambitions beyond enterprise tooling. Expect a preview at DevDay next week that could reshape how content and commerce tools plug into ChatGPT.
Supply Chain & Ops
5 storiesNew report: Just a third of container shipping on-time
Schedule reliability on Far East-Europe lanes has slid toward pandemic-era lows. Planners should widen buffer windows and revisit safety stock assumptions built around post-pandemic normalcy.
Record-low Rhine levels choke barge traffic, stress ground transport
Low water is pushing volume off barges and onto trucks and rail across central Europe, adding cost and lead time risk for anyone sourcing through German inland routes this quarter.
Electronics manufacturers fret over extreme heat disruptions
Heat is delaying supplier deliveries and cutting productivity at electronics plants. Climate risk belongs in supplier scorecards alongside price and quality now.
How a team learns to trust a quote it did not build
Cargo.one's data shows AI-generated freight quotes now go out untouched four times out of five, but getting there took a deliberate trust-building process, not just better models. A useful playbook for any ops team rolling out automated pricing.
Toyota orders workers to train humanoid robots but says humans won't be replaced
Toyota is putting factory staff to work training robots ahead of a planned rollout of 400,000 units, a preview of how manufacturers plan to blend human oversight with robotic labor rather than swap one for the other outright.
Deals & Market
5 storiesSnorkel AI triples valuation to $3.5B as demand for AI training data booms
A $350 million round on tripled valuation shows training data curation is still a scarce, high-margin business even as model prices fall. Worth watching if you're sourcing labeled data for internal models.
HutchWatch: The $23bn port deal that nobody can close
Eighteen months after CK Hutchison agreed to sell 43 ports to BlackRock and MSC, the deal still hasn't closed. A reminder that terminal control consolidation is moving far slower than headlines suggested.
Union Pacific & Norfolk Southern combo 'gains additional momentum'
The Surface Transportation Board rejected a bid to kill the merger outright, keeping the biggest rail consolidation in years alive. Worth tracking for anyone planning intermodal routing.
Nscale's IPO will test Wall Street's appetite for concentrated AI bets once again
A data center developer that depends on Microsoft and Anthropic for most of its revenue going public is a live test of how much concentration risk investors will tolerate in AI infrastructure plays.
Coca-Cola to spend $10B on US manufacturing by 2030
A decade-scale bet on domestic production and distribution capacity signals large CPGs are still willing to fund physical footprint even amid AI capex headlines dominating budgets elsewhere.
Research & Frontier
4 storiesOpenAI says its internal model solved over 100 long-standing math problems after just a month of training
Mathematicians are pushing back on OpenAI's pace and self-reported results, and the company's own advisory group doesn't get to review research speed. A governance gap worth watching as these claims get used for marketing.
Deep Reinforcement Learning on Item-Compatibility Graphs for One-Dimensional Bin Packing
A niche paper but directly applicable to warehouse and logistics packing problems. Learned methods are starting to handle constraint-heavy bin packing beyond the toy 2D and 3D cases most research covers.
Pentagon says overreliance on AI contributed to missile strike on Iran school
A documented case of AI-assisted targeting contributing to a civilian strike is the kind of failure analysis every operations leader deploying decision-support AI should read before trusting model outputs over human judgment.
Agreement Overstates Evidence: Error Dependence in LLM Judge Consensus
Multiple AI judges agreeing doesn't mean they're right, they often share training biases and make correlated mistakes. A caution for anyone using LLM panels to grade vendor outputs or automate QA.
Org & AI Architecture
5 stories'We're already fighting yesterday's battle': Greece's prime minister gets candid about AI
A rare admission from a sitting head of government that no policy framework is ready for what AI deployment actually looks like. Worth reading for anyone waiting on regulatory clarity before scaling AI use.
OpenAI calls for international standards on AI that could improve itself
OpenAI wants global oversight rules for recursive self-improvement before it happens rather than after, a notable shift toward proactive governance talk from a company usually criticized for moving fast.
Lawsuit demands OpenAI pay for new school after ChatGPT used in shooting
British Columbia's suit and demand for chatbot logs raises the stakes on what companies must disclose when their tools are implicated in real-world harm. A liability question every enterprise deploying chat assistants should watch.
Meta admits Muse's likeness to OpenClaw isn't a coincidence
Meta acknowledging its assistant was heavily inspired by a competitor down to workspace filenames is an awkward admission that raises IP questions worth watching if you're evaluating vendor AI assistants for originality or licensing risk.
Drivers at DHL Express delivery vendors in Dallas join Teamsters union
Unionization spreading through outsourced delivery vendors is a labor signal worth tracking alongside automation rollouts. Carriers leaning on contracted delivery networks should expect more organizing activity.