Waymo's move into Spain leads the AI briefing for October 20: in the past 24 hours, Waymo said it aims to carry its first riders in Madrid in 2027, DeepSeek's updated V4.1 Flash climbed to the top tier of open models on two leaderboards, Reuters reported OpenAI is in talks for roughly $30 billion in new funding, ModelBest squeezed a 35B-class MoE model onto a phone, and Meta's Muse Wonder took sixth place on a web development chart.
Waymo lands in Madrid: manual driving first, riders in 2027
Waymo announced its move into Spain on its official blog on October 20, 2026, starting with the Madrid region and aiming to welcome its first riders in 2027. In the coming days its electric Jaguar I-PACE vehicles will appear on local streets, beginning with manual driving before trained local autonomous specialists run technical validation. Waymo says it has completed more than 20 million trips and over 400 million fully autonomous kilometres, with crashes involving serious injury or worse running 95% below human drivers in its operating cities. For Europe, it is another step by a US self-driving company into a major EU capital.
DeepSeek V4.1 Flash update climbs the charts: No. 1 open model on Agent Arena, 45 on the Intelligence Index
Arena announced on October 19, 2026 that DeepSeek-V4.1-Flash-20261017 (Max) is now the top open model on Agent Arena and No. 11 overall, based on more than 7,500 real sessions, with a net improvement of 5.1%, a median task cost of about $0.20 and an 11.5% gain in confirmed success. The same day, Artificial Analysis scored the release 45 on its Intelligence Index, up 6 points, tied as the second open-weight model alongside GLM-5.3. Same price, higher scores: cheaper models are taking on more real work.
OpenAI reported to be raising $30 billion at a valuation above $1.4 trillion
According to Reuters reporting on October 20, 2026, OpenAI is in talks to raise about $30 billion at a valuation above $1.4 trillion, moving ahead of a planned Anthropic listing. That figure is at least 64% above its roughly $852 billion level in March. OpenAI has not officially confirmed the talks, but if the round closes, the funding race among frontier labs escalates again, and so do the prices of compute and talent.
ModelBest's LES scheduling runs a 35B MoE on a phone in real time
ModelBest's MiniCPM team explained its Lookahead Expert Scheduling (LES) technique on its official WeChat account on October 20, 2026, saying it lets MiniCPM-V 4.7 run a 35B-A3B MoE model on-device in real time without quality loss, decoding at over 20 tokens per second on an iPhone 18 Pro Max. The hard part of on-device MoE is expert-scheduling latency, not total parameters; if this approach holds up, the ceiling for local multimodal assistants on phones rises.
Meta Muse Wonder debuts at No. 6 on the WebDev chart with 1,726 points
Arena announced on October 19, 2026 that Meta's Muse Wonder Full (Max) debuted at No. 6 on the Code Arena WebDev leaderboard with 1,726 points, lifting Meta to third place among labs on that chart, a spot previously held by Google. The Muse family is Meta's new generation of foundation models, and phase one of its Prometheus supercluster was built to train it; the ranking shows Meta's frontier models now competing head-on in hard scenarios such as coding.
