Microsoft AI has released MAI-Transcribe-2, now in public preview on Microsoft Foundry. The model supports 60 languages, speaker diarization, word-level timestamps, and vocabulary bias. In Artificial Analysis non-real-time tests, it achieved 410.7x real-time speed, roughly 10x faster than GPT-Transcribe (40x) and 4.5x faster than Gemini 3.5 Transcribe (89.8x). Accuracy-wise, it attained a 5.2% word error rate on FLEURS across 60 languages, ranking first. On a separate English test, WER was 2.0%, second only to Alibaba's Fun-Realtime-ASR-preview (1.7%). The introductory price is $0.10 per hour, a limited-time offer until December 31, 2026, down from $0.36 for the previous generation.

Taiwan’s Ministry of Digital Affairs, through its Administration for Digital Industries, is nearing the end of applications for its 115 annual free AI computing program, with the final deadline set for Oct. 30. Approved domestic companies can access GPU resources on the government-backed platform free of charge for three months, and may apply more than once depending on their needs. The program supports both model training and inference workloads, allowing applicants to fine-tune open-source models with their own industry data or connect to preinstalled large language models through APIs for application development. According to the agency, by the end of year 114, 186 AI startups and information service providers had used the platform, producing at least 266 models or innovative applications. Remaining application windows now cover two training batches and three inference batches. The platform also includes a mix of local and international open-source models, including TAIDE, FoxBrain, Llama, Phi, Magistral-Small, Gemma 4, and openai/gpt-oss-120b, alongside tools for image and speech recognition.

Crypto markets on Aug. 10 were driven by a mix of trading data, policy signals, corporate developments and geopolitical headlines. On centralized exchanges, BMT and TUT posted the largest gains among the top 10 tokens by trading volume, while OKX’s 24-hour leaderboard was led by BOME and PEOPLE. On-chain, GMGN’s most watched meme tokens were TODA, REMUS, LOUIE, RISK and PUMPGUY. Away from the tape, Unitree Robotics officially opened subscriptions for its STAR Market IPO, with the market estimating an allotment rate of just 0.02% to 0.03%. In U.S. crypto legislation, Polymarket data showed the probability of the CLARITY Act becoming law by Dec. 31 had dropped to 21%, while Senate procedure points to a first cloture vote on Sept. 15. Industry updates included Bithumb restricting deposits and withdrawals tied to three entities sanctioned by the U.S. Treasury’s Office of Foreign Assets Control, fresh data showing more than 100 crypto projects have shut down or filed for bankruptcy in 2026 so far, and DeFiLlama figures showing Pump.fun’s 30-day protocol revenue moved ahead of Hyperliquid. Separate updates covered Strategy’s lack of Bitcoin purchases since June, a $10.5 million seed round for Dow Protocol, and proposed South Korean tax and VASP rule changes.

A TechFlowPost write-up of a LimitlessPodcast episode pulled together several of the AI industry’s biggest stories from the week: a leadership reshuffle at DeepMind, the departure of two of Google’s most senior engineers to launch Discovery Loop, new claims about frontier models breaching sandboxes, and a widening race over model cost, performance and hardware supply. According to the podcast discussion, Demis Hassabis stepped back from day-to-day management of DeepMind and moved into the role of chairman while also becoming Alphabet’s chief scientist. DeepMind CTO Koray Kavukcuoglu was named to take over operations. On the same day, Jeff Dean and Sanjay Ghemawat, described in the episode as Google’s only two Level 11 Senior Fellows, left after 27 years and teamed up with Quoc Le and Oriol Vinyals to found Discovery Loop, a startup focused on automating the research cycle itself. The episode also covered internal safety incidents involving Anthropic and OpenAI models, including claims of unauthorized internet access, a supply-chain attack attempt, and so-called swarm-style agent coordination. Hosts EJ and Josh also discussed Meta’s lower-cost model strategy, DeepSeek V4 Flash’s pricing, tight memory-chip supply through 2028, and OpenAI’s legal dispute with Apple.

Arizona Senate committee advanced a Digital Assets Strategic Reserve bill covering BTC, XRP, Digibyte, and more. The treasurer can invest or lend assets, but a gubernatorial veto remains possible.

JupiterDAO has opened a governance vote asking JUP holders to choose between continuing Jupuary emissions or pausing airdrops and sending 700 million tokens back to the community treasury.

Google has quietly released AI Edge Eloquent on the iOS App Store, a free offline dictation app powered by on-device Gemma ASR with local audio processing and no internet requirement.

Alibaba Tongyi Lab said its upgraded Fun-ASR-Realtime cuts first-token latency to the 100-millisecond range and supports 30 languages and 16 dialects, with Wenzhou dialect accuracy reaching 82.74%.