Morning edition Coverage: overnight, since the evening brief (~19:40 → 08:20 CEST)

What matters

Anthropic’s Haiku 5.5 slashes small-model prices as OpenAI rolls GPT-6 out to all ChatGPT users; Apollo software pioneer Margaret Hamilton has died.


Filter by topic

12 stories

Lead
  • AI
  • Dev

Anthropic launches Claude Haiku 5.5: its fastest model, at about a quarter of Haiku 4.5’s running cost

Claude Haiku 5.5 announcement graphic
Anthropic
  • Haiku 5.5 is the third model of the 5.5 generation, two weeks after Opus 5.5. Anthropic pitches it for high-volume work (summaries, classification, compaction) and as a subagent under Opus or Sonnet on coding jobs.
  • List price drops to $0.10 per million input tokens and $0.50 output for prompts up to 100k tokens, 90% below Haiku 4.5’s $1/$5. Longer prompts get 50% off. Anthropic’s “~75% cheaper on average” figure accounts for a new tokenizer (SiliconANGLE).
  • First Haiku with an adjustable effort setting. Anthropic’s own benchmarks: 72.4% on OSWorld 2.1 (offline subset) vs 48.9% for OpenAI’s same-priced GPT-6 Luna; 39.2% on Terminal-Bench 4.0 vs 16.4%.
Why it matters — Small-model pricing just fell by an order of magnitude on the Claude side, matching OpenAI’s Luna. For agent pipelines that fan out to many subagent calls, that changes the cost math more than any flagship launch.
  • AI

OpenAI brings GPT-6 to every ChatGPT tier, with “Intelligent UI” answers built from live widgets

Examples of ChatGPT’s Intelligent UI answers with interactive elements
MacRumors
  • GPT-6 Sol rolled out on Wednesday to Plus, Pro, Business and Enterprise; Free and Go users get GPT-6 Luna starting today (Thursday). GPT-6 is limited to the Chat tab for now (MacRumors).
  • Intelligent UI lets answers include tappable buttons, forms, charts, maps, checklists and small generated tools such as a bill splitter or a recipe that rescales to guest count.
Why it matters — The chat window is turning into an app surface. That matters for anyone building on top of ChatGPT, and it lands the same evening Anthropic undercut it on small-model price.
  • Science
  • Tech

Margaret Hamilton, who led the Apollo flight software team, has died at 90

Portrait of Margaret Hamilton
MIT News
  • MIT announced on Wednesday that she died on 30 September. She led the software engineering team at MIT’s Instrumentation Lab during NASA’s Apollo program.
  • She worked at MIT from 1959 to the mid-1970s, wrote more than 130 publications and helped establish software engineering as its own discipline, before founding companies and serving as CEO.
  • AI
  • Science

Mathematicians push back on OpenAI’s 722-paper math dump; Terence Tao reposts a call to stop working with OpenAI

Photo of mathematician Terence Tao sitting cross-legged on a wood floor, partly in sun and partly in shadow, shot from above.
Fortune
  • Tao’s blog reposted a statement from the Association for Human Mathematics calling the 6 Oct release “not a demonstration of scholarship, but a demonstration of power” and urging mathematicians to discontinue work with OpenAI.
  • Fortune counts full or partial solutions to more than 370 open problems. Toronto’s Dan Litt calls it “great for mathematics” but worries about funding and young researchers; NYU’s Tristan Buckmaster says OpenAI hasn’t done due diligence on whether it built on others’ work.
Why it matters — This is a material turn from yesterday’s release story: the field’s most prominent mathematician amplifying a boycott call, while others welcome the results. Verification and credit are now the fight.
  • Markets

Fed minutes point to one more hike by year end; a 10-year auction clears at the highest yield since 2000

  • Minutes of the September meeting: “most participants” saw another rate increase as likely appropriate by year end; Sharecast reads it as many happy to wait until December. WSJ: no urgency for October.
  • The $39bn 10-year auction priced at 5.300%, the highest at an auction since late 2000. The 10- and 30-year yields closed at 5.361% and 5.730%, highest since mid-2002.
Why it matters — Follow-up to last night’s bond sell-off: the Fed isn’t rushing, but the market is still repricing long-term debt higher, and that is feeding straight into mortgages.
  • Tech
  • Markets

Samsung guides to an $80bn quarterly operating profit on AI memory demand

Samsung logo at an exhibition stand
CNBC / Getty Images
  • Preliminary Q3 operating profit of 107.4 trillion won (~$80bn), the first time above 100 trillion won and about 782% above a year ago (12.17 trillion). Revenue about 195 trillion won, up ~127%.
  • Benzinga/Reuters put the LSEG SmartEstimate at 106.1 trillion won. Shares still slipped about 0.7% in Seoul; eToro’s Josh Gilbert says the market “doesn’t hand out points for effort.”
  • World

Trump says he no longer wants an Iran deal as the US readies possible strikes; Houthi attacks on Saudi airports kill three

Scene after an attack at a Saudi airport
Al Jazeera / Reuters
  • At a San Antonio rally, Trump said a deal “isn’t really something that I want to do.” NBC reports his team has discussed resuming large-scale operations in the coming weeks; Axios says that could mean “massive bombing” of energy, infrastructure and nuclear targets.
  • Houthi-claimed attacks hit Abha and Riyadh’s King Khalid airports on Tuesday and Wednesday: three foreign residents killed, 36 injured, per Saudi civil aviation.
  • Germany

Coalition committee ends quietly after four hours; Merz floats a “false flag” theory over the AfD Landtag vote

  • Union and SPD leaders met at the Chancellery for about four hours, the first coalition committee in over three months. Thorsten Frei leaving: “Gut war’s.” Results are due in writing this morning, with no press conference (ZDF).
  • Topics were the July reform package (pensions, care, labour market, taxes) and the 2027 budget. Merz already said he doesn’t expect the package done by Christmas, hoping for spring.
Brief
  • Dev notes

Dev notes

  • Playwright 1.64 adds page.webmcp / frame.webmcp to list and call tools a page registers via the experimental WebMCP browser API; Playwright MCP now offers them to agents by default (--no-webmcp to opt out). The release is flagged with a breaking change. Playwright v1.64.0

Morning edition 12 stories 49 sources

Throwaway briefing page. No accounts, no tracking. Optional story votes stay in your browser (and GET-relay only when PUBLIC_VOTE_URL is configured). Source links open in a new tab.