Daily
AI Diary
A brief daily diary of what is changing in artificial intelligence.
September 27, 2026
AI Diary — 27 September 2026
Today was quiet. No new model announcements, confirmed price changes, or movement in the independent rankings. The picture we shared yesterday remains exactly the same today — that is not news, but it is worth knowing.
September 26, 2026
AI Diary — 26 September 2026
The largest independent leaderboard comparing AI models was updated yesterday after twelve days of silence, and a new name took the top spot. On the same day, Anthropic announced that it would now charge for some requests blocked by its own safety filters. Taken together, these developments paint a clear picture: the field is both accelerating and rewriting its own rules.
September 25, 2026
AI Diary — 25 September 2026
Yesterday brought no new race for the "smartest model." Google made a technology available to everyone, while several companies fixed and sped up their existing tools. It was a quiet day of genuine progress.
September 24, 2026
AI Diary — 24 September 2026
Yesterday was busy, though not noisy. Google announced three separate products in a single day, Anthropic made headlines with a scientific discovery, and NVIDIA released an open-source audio model. None of this was about the race for the "best model" — each offered a concrete feature ready to use today.
September 20, 2026
AI Diary — 20 September 2026
Yesterday's most striking news came not from a laboratory but from a hospital: Alibaba open-sourced an imaging model, and its results were published in a peer-reviewed journal. On the same day, Claude Code received an extensive security update, while small open-source models continued to attract interest. Healthcare, security, and accessibility — all three came together in a single day.
September 18, 2026
AI Diary — 18 September 2026
One of the year's most tangible developments came from locally run models: a powerful AI model with 27 billion parameters now fits on an ordinary laptop. On the same day, Anthropic made moves on two separate products — one expands access to expertise, while the other changes how Claude works. It was a quiet but eventful day.
September 16, 2026
AI Diary — 16 September 2026
Yesterday, Google announced two audio models at once, but they did not attract equal attention. One genuinely took first place in an independent ranking — seeing the margin makes you realize just how tight the race is. The other had been out for a month without anyone noticing. On the coding side, meanwhile, there is a small but real security improvement.
September 13, 2026
AI Diary — 13 September 2026
Yesterday's biggest talking point was not a model announcement. Anthropic's founder wrote that the industry needed to slow itself down, and within hours the heads of two rival companies said they agreed. On the same day, Anthropic also announced a change to the usage limits for its coding tool, Claude Code, starting tomorrow. It was a quiet day for new models.
September 12, 2026
AI Diary — 12 September 2026
There were no major model announcements yesterday. Most developments involved refinements to existing tools: a music model opened the door to commercial use, a coding tool patched a security vulnerability, and a small shift on a ranking site offered an example of something that could be read as "the model has regressed" but was not.
September 11, 2026
AI Diary — 11 September 2026
Yesterday was a busy day for open source: DeepSeek released a new model under the MIT license — meaning it is free and available for commercial use — but its own account and the independent evaluation did not tell the same story. Meanwhile, Claude Code received a series of security patches, and OpenAI introduced a new voice model.
September 10, 2026
AI Diary — 10 September 2026
There was no single big story yesterday, but there were several complementary developments: a developer and an independent testing organization reported different numbers for the same comparison, a music generation tool signed a licensing agreement with major record labels, and a small but powerful open-source model was released.
September 9, 2026
AI Diary — 9 September 2026
Yesterday brought two separate claims in fluid dynamics, both supported by AI. On one side are two independent mathematicians, on the other OpenAI — with a dispute between them over who got there first. There was also a new personal AI agent, two image models, and a massive genetic database made freely available.
September 8, 2026
AI Diary — 8 September 2026
There is no major model release today either — but a number we reported yesterday turned out not to be what it seemed. An independent ranking system changed its own metric, a reminder of how carefully we need to read headlines such as "approaching the top of the rankings." There was also a new open-source model, small enough to run even on a phone.
September 7, 2026
AI Diary — 7 September 2026
Yesterday brought no major model releases or acquisition news; the day's developments were more subtle. One model configuration approached the top of an independent ranking, an open-source tool began communicating with a new platform, and a company shut down an old connection method entirely. All three are small but real changes.
September 6, 2026
AI Diary — 6 September 2026
The largest platform for sharing AI models is changing hands: NVIDIA is acquiring Hugging Face. There is also news from mathematics — an AI system has made the proof of a 370-year-old theorem verifiable by computer. Both stories raise the same questions from different angles: who controls the AI ecosystem, and what do these systems actually do well?