A 1941 message, an open image model and Paris tenth

Alibaba has open-sourced Qwen-Image-2.1: a single model that handles both text-to-image generation and image editing, with 7 billion parameters in its visual generation component (32 single-stream DiT layers). The team highlights native transparency — generating and editing images with an alpha channel, extracting a subject from a photo — a capability previously handled by a dedicated model, Qwen-Image-Layered, released in December 2025.

For editing, the model takes up to ten reference images, supports local edits, and is designed to preserve people and products in an image. Efficiency comes from a mixed-granularity attention architecture and KV cache reuse, treating input images as static context computed once. Worth knowing before reading anything into a ranking: the published comparisons are from Qwen-Image-Bench, the team's own benchmark.

↑ Back to the summary

On 10 July 1941, a German Army radio message asked for a route of march and an immediate reply. It stayed unsolved in the CryptoCellar/Weierud corpus, still listed as “Fails” as of 14 September 2026. On 14 and 15 September, an investigation run with AI agents recovered the machine settings and the text: “Please specify the route of march. I am in Rosenow, Rosenow. Immediate reply by radio.”

Brute force was not the answer: the message body is only 82 letters, the machine allows roughly 159 quintillion daily configurations, and the two surviving copies — a published transcription and a faint handwritten facsimile — disagree at several positions, where a single wrong letter is enough to make a correct key look wrong. The report is explicit about what it does not claim: this is neither the historical breaking of Enigma nor a corpus-wide breakthrough. It also notes that its logs give no complete total of reasoning hours, human or machine — and it puts no figure on the compute spent.

↑ Back to the summary

According to a study by the fund Asgard Capital and consultancy Roland Berger, Paris is the tenth city worldwide by number of AI startups, with 73 — exactly as many as ninth-placed Los Angeles. Ahead of them: San Francisco and Silicon Valley (596), London (211), Tel Aviv (189), New York (180), Beijing (150), Boston (102), Tokyo (99), Shanghai (77).

Paris remains the European Union's leading AI city, well behind London, and holds 73 of the more than 1,000 AI startups France counts — Mistral, but also Dust and Photoroom. The study counts companies, not headcount or funding: it describes ecosystem density, not economic weight. The authors also note that the worldwide total is itself a wide estimate, between 35,000 and 90,000 AI startups.

↑ Back to the summary