Afternoon edition — Thursday 29 October 2026

Mistral on 29 October published the weights of an improved Mistral Large 4, nicknamed Le Chonk, which it first offered as a public preview earlier this month. Mistral describes it as a 1-trillion-parameter, natively multimodal model trained from scratch in its own European data centres. The weights are on Hugging Face, and the model is also offered through Mistral's API and OpenRouter.

Mistral says three more weeks of reinforcement learning made it the strongest open-weight coding model. It reports 48% on the Artificial Analysis Coding Agent Index, level with Zhipu's GLM-5.3 while using 47% fewer tokens per task, and 41% on Terminal-Bench 4.0. On security, it reports 51% on the Artificial Analysis Cyber Index and a 96.6% average refusal rate on malicious cyber prompts.

The licence is a modified MIT licence: use, fine-tuning and commercial use are free, but companies selling the model as a service need a separate licence. API pricing is $0.70 per million input tokens and $2.10 per million output tokens. Mistral also says enterprise partners such as Orange Cyberdefense get preview access to a specialised variant for penetration testing and red-teaming, under vetting and contracts.

Speech-AI company Deepgram is acquiring Daily, the company behind Pipecat, a widely used open-source framework for building voice agents, the two companies announced on 29 October. Terms were not disclosed. Daily also runs real-time audio and video network infrastructure, and its team is joining Deepgram.

Both companies say Pipecat will stay open source under its existing BSD licence and vendor-neutral. They are setting up an independent Pipecat Foundation which, according to Daily, will be run by a steering committee that includes companies competing with Deepgram.

At its first Speak conference, Deepgram also introduced the Real-Time Context Standard, an open protocol written with the Pipecat team. It carries cues such as a whisper or an unclear word from one model to the next, so that a voice agent can ask for clarification or match its tone.

Alphabet shares fell 13% in early New York trading on Thursday 29 October, CNBC reported, after third-quarter results released after the market closed on Wednesday. CNBC noted that a close more than 9.77% lower would be the company's worst reaction to earnings since it went public in 2004.

Revenue grew 25% to $127.58 billion but missed analysts' forecasts, as did adjusted earnings of $2.67 a share. Free cash flow was minus $9.07 billion, against the minus $2.70 billion analysts expected. Capital spending reached $51.36 billion in the quarter, and Alphabet kept its 2026 forecast of $195 billion to $205 billion, saying spending will increase significantly in 2027.

Search advertising grew 12% to $63.35 billion, below expectations. On the earnings call, business chief Philipp Schindler said the growth of AI Mode queries had an impact because they currently earn less than traditional search. Google Cloud revenue more than doubled to $30.7 billion, according to The Decoder.

Microsoft, which also reported on Wednesday, was little changed. Its capital spending including finance leases was $45.2 billion, below forecasts, and its free cash flow stayed positive at $18.63 billion.

Microsoft is selling at least $25 billion of bonds in its first US high-grade debt sale since 2017, Bloomberg reported on 29 October, citing a person with knowledge of the matter. The deal has as many as 10 parts, with maturities from 2 to 40 years; Wells Fargo, JPMorgan Chase and UBS are managing it, according to The Next Web.

Microsoft last raised $17 billion in a US bond sale in January 2017. It is one of only two US companies with a AAA credit rating. The sale follows results in which Microsoft's spending on data centres and equipment, including leases, reached $45.2 billion in the quarter to September and free cash flow fell 27%. Chief financial officer Amy Hood told analysts: 'We still have demand that exceeds supply.'

US intelligence agencies concluded that Alice Huang (Huang Le), a Chinese citizen and senior executive at the Singapore-based data-centre company Megaspeed, was passing information to Beijing, The New York Times reported on 29 October, citing people briefed on classified surveillance. The surveillance showed she had developed a close relationship with Nvidia chief executive Jensen Huang.

According to the report, as summarised by The Decoder, President Trump was briefed about a year ago and took no action, and Jensen Huang was not warned; officials believed he was unaware. US authorities already suspected Megaspeed of helping Chinese companies obtain Nvidia chips, and the Times reports that a Malaysian sister company bought at least $4.8 billion of AI hardware with Nvidia chips between June 2024 and February 2026.

Nvidia told The Verge she is one of the thousands of customer executives Huang meets each year, and said Megaspeed does not receive sensitive information. Megaspeed has announced an internal investigation, and the Chinese embassy called the allegations groundless smears.