Thursday, October 8, 2026

AidenIt's Thursday, October 8, 2026. Anthropic released Claude Haiku 5.5, the first Haiku with an adjustable effort setting.

AidenI'm Aiden, and this is DailyChat, from Silicon Valley. Jonah and I are AI characters, and the facts come from the sources we link.

AidenJonah is our AI products analyst. Jonah, what's available today?

JonahIt's live now on the Claude Platform, Amazon Web Services, Google Cloud and Microsoft Azure. Input costs $0.10 per 1M tokens and output $0.50, for requests up to 100,000 tokens.

AidenHow does that compare with the last Haiku?

JonahHaiku 4.5 was $1 in and $5 out. On the OSWorld 2.1 agent test, Anthropic reports 72.4% for Haiku 5.5, against 15.7% for Haiku 4.5. That's Anthropic's own score, on an offline subset of the test. It also says the new tokenizer uses slightly more tokens per task.

AidenSo, for you: if you build with Claude, test it on your own tasks first, and compare the cost per task as well as the price per token.

AidenDailyChat is AI-generated, with AI characters. Not professional advice. See you tomorrow.

AidenIt's Thursday, October 8, 2026. GPT-6 is rolling out in ChatGPT's Chat tab, with answers that can include interactive charts, forms and buttons, MacRumors and TestingCatalog report.

AidenI'm Aiden, and this is DailyChat, from Silicon Valley. Jonah and I are AI characters, and the facts come from the sources we link.

AidenJonah is our AI products analyst. Jonah, who gets it, and when?

JonahPaid users first. Plus, Pro, Business and Enterprise get GPT-6 Sol from October 7. Free and Go users get GPT-6 Luna a day later.

AidenAnd what is the Intelligent UI?

JonahAnswers can include charts, forms, tappable buttons and maps. MacRumors says it can even build small tools, like a bill splitter. It applies only to the Chat tab. The models for Work and Codex are not changing. All of this comes from those 2 reports.

AidenSo, for you: look in the Chat tab over the next 2 days. If you use ChatGPT for work or code, those models aren't changing.

AidenDailyChat is AI-generated, with AI characters. Not professional advice. See you tomorrow.

AidenIt's Thursday, October 8, 2026. OpenAI has withdrawn 3 of the math manuscripts it posted a day earlier, after finding a sign error.

AidenI'm Aiden, and this is DailyChat, from Silicon Valley. Mira and I are AI characters, and the facts come from the sources we link.

AidenMira is our AI research analyst. Mira, what went wrong?

MiraOpenAI posted a repository of 719 math manuscripts. On October 7, its own history page says it withdrew 3 of them.

AidenWhy those 3?

MiraA sign error. OpenAI says it breaks a cancellation argument in one paper. 2 other papers use that paper's construction, so they fall with it. Picture a building with one wrong beam, and 2 floors resting on it.

AidenIs that all that changed?

MiraNo. OpenAI says it fixed 14 other manuscripts, with proof repairs, corrected statements and clearer hypotheses. Another 13 were updated to cite the revised versions. The withdrawn papers carry notices explaining the gap and linking the archived manuscripts.

AidenDoes OpenAI say how many results are checked?

MiraIt reports that 300 of 719 top-line results are formalized, about 42%. That's OpenAI's own count, and it means most results are not in that group. The page doesn't say how the sign error was found.

AidenAnd what do mathematicians say?

MiraThe Association for Human Mathematics published a statement. It says mathematicians didn't ask for this work, and it urges them to stop working with OpenAI. That is one group's position. For the evidence, the useful part is that the withdrawal is public, with the reasons written down.

AidenSo, for you: treat any single one of these results as unconfirmed until it has been checked outside OpenAI, and follow the notices in the sources we link.

AidenDailyChat is AI-generated. Aiden and Mira are AI characters, not real experts. Facts come from the sources we link. Not professional advice. See you tomorrow.

AidenIt's Thursday, October 8, 2026. OpenAI says it has disrupted its first Category 5 influence operation, one of 2 it recently banned.

AidenI'm Aiden, and this is DailyChat, from Silicon Valley. Henry and I are AI characters, and the facts come from the sources we link.

AidenHenry is our AI safety analyst. Henry, what did OpenAI find?

HenryOpenAI says it banned ChatGPT accounts behind 2 operations built on false fronts: entities that hide who is behind a message. One is Russia-origin, called Dark Clark. The other is Iran-origin, called Bogus Bylines.

AidenWhat did each one do?

HenryBogus Bylines used 7 fake journalist personas to pitch long articles to small and medium online outlets. It also generated batches of social media comments. Dark Clark ran a self-described research platform in Latin America. OpenAI says the staff there appear not to have known they worked for a Russian group.

AidenAnd the Category 5?

HenryThat's the IO Breakout Scale, from 1 to 6. OpenAI rates Dark Clark Category 5, the first it has disrupted, and Bogus Bylines Category 4. Both landed content in mainstream media. OpenAI sees that pattern across the 30 operations it has exposed in 2.5 years. The rating is OpenAI's own assessment of potential reach.

AidenSo did AI do the work?

HenryLess than you might think. OpenAI says these operations closely resembled influence operations from before AI, and that AI made some workflows easier. For Dark Clark, it mostly saw ChatGPT used to write internal reports, with little campaign content. It also says both groups used questionable methods to exaggerate their own effectiveness.

AidenWhat's confirmed, then?

HenryThe ban, and OpenAI's description of what the accounts did. OpenAI says links between Dark Clark and a Russian group called Politology come from outside researchers. I wouldn't read more into it than that.

AidenSo, for you: when a polished article or think tank report reaches you, check who publishes it and who is behind it. Editors can read OpenAI's full report in the sources we link.

AidenDailyChat is AI-generated. Aiden and Henry are AI characters, not real experts. Facts come from the sources we link. Not professional advice. See you tomorrow.