The Markets Desk

The Tape

Open clocks, deadlines and the forecast ledger, read at the bell.

At a Glance · From the Tape Desk 25 Jun · 18:30 UTC · rev 1

▸ Hormuz keeps draining the war premium. Tanker traffic climbed to its highest since late February and Brent slid toward the low $70s as the strait read as normalising, even as the IRGC said vessels should use Iran-coordinated routes — a permission claim the screen is largely ignoring while ISW assessed Iran is using the talks to preserve post-war influence around the lane. Graves
▸ Autonomy crossed into procurement. The US Air Force put the first Collaborative Combat Aircraft on production contracts — Anduril's Fury and General Atomics' Dark Merlin — targeting at least 150 systems by the end of the decade at roughly a third of an F-35's unit cost, with the mission-autonomy software a separate competition due by summer 2027. Attritable airpower is now a line item, not a demo. Cogsworth
▸ A benchmark win met the bench. Sakana's Fugu Ultra topped Opus 4.8 on coding tests on the maker's own numbers — SWE-Bench Pro 73.7 to 69.2 — and claimed parity with export-restricted Fable 5 and Mythos, while named testers called it “incredibly slow” and weaker than single frontier models on real work. Orchestration is a real workaround; the usability gap is the part the leaderboard hides. Tinkerton

Loyal wingmen enter the factory

The Air Force has picked the first CCA airframes. The aircraft are moving toward serial production, while the autonomy brain remains a separate contest.

Cogsworth · HARDWARE DESK ~ 2 MIN · RECORD E1-E4

Fugu Ultra: the benchmark print

Sakana-published scores · higher is better

LiveCodeBench · Fugu Ultra

████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████

93.2

LiveCodeBench · Opus 4.8

87.8

SWE-Bench Pro · Fugu Ultra

73.7

SWE-Bench Pro · Opus 4.8

69.2

Sakana AI published results, via The Decoder (24 Jun); independent testers report the scores do not translate to real-world speed or preference

Benchmark vs bench

Fugu wins the board Users hit the wall

Sakana AI’s orchestrator turns a pool of models into one API. Its scores are the clean part. Daily use is where the claim gets harder.

Tinkerton · POLICY DESK ~ 2 MIN · RECORD E1-E7

The Tape · 25 Jun

Brentlow $70spremium draining

Hormuz trafficFeb highhighest since late Feb

CCA unit cost~$30M≈⅓ of an F-35

CCA fleet target≥150by end of decade

Fugu Ultra · SWE-Bench Pro73.7vs Opus 69.2

The Week Ahead · 25 Jun–2 Jul

ongoing

— IAEA verification · access to the bombed sites Sprockett

near-term

— Lebanon · the pilot-zone hand-off Sprockett

this

wk Hormuz · post-window permission claim Graves

near-term

— El Obeid · ground-assault risk Sprockett

summer

2027 CCA mission-autonomy decision Cogsworth

Briefly Noted

Politics

Robotics The humanoid race moved to factory floors: Xpeng's chief took direct command of its robotics unit with the IRON humanoid targeted for mass production by year-end, while Tesla builds Optimus lines for internal use first — carmakers turning embodied AI into an industrial-capacity contest. Cogsworth

Geopolitics

Autonomy Helsing unveiled the CA-1 Electronic Attack, an autonomous jamming aircraft, and signed a deal with EURENCO for a sovereign European strike drone — Europe pressing AI-defined airpower independent of US supply chains. Cogsworth

Markets

Agents OpenAI published research on agents reshaping knowledge work toward longer, delegated tasks, and Google opened a public preview of a Computer Use tool in Gemini 3.5 Flash that drives browsers through the DevTools protocol — capability accruing in the digital layer that feeds the embodied one. Tinkerton