Lookonchain APP

App Store
Feeds Articles

Microsoft’s new ASR model is 10 times faster than GPT-Transcribe, costing only $0.1 per hour.

1 hours ago

Beating AI News reports: Microsoft AI has launched MAI-Transcribe-2, now available in public preview on Microsoft Foundry. The new model supports 60 languages, adding speaker diarization, word-level timestamps, and domain-specific term biasing. Developers can pre-load names, drug terms, or industry jargon to reduce mishearing of rare words. The most notable improvement is speed. In non-real-time transcription tests by Artificial Analysis, MAI-Transcribe-2 achieved approximately 410.7x real-time speed, meaning a 1-hour audio recording can theoretically be processed in around 9 seconds. GPT-Transcribe clocked in at around 40x, making MAI-Transcribe-2 roughly 10 times faster; Gemini 3.5 Transcribe hit about 89.8x. Accuracy is also among the top tier. In FLEURS’ 60-language test, it posted an average Word Error Rate (WER, lower is better) of 5.2%, ranking first; in another non-real-time English test by Artificial Analysis, it had a WER of 2.0%, placing second, only trailing Alibaba’s Fun-Realtime-ASR-preview at 1.7%. Pricing is currently set at $0.10 per hour, down from $0.36 for the previous MAI-Transcribe-1.5. However, this is a limited-time offer valid only through December 31, 2026; the official full price has not been announced yet.

Source
Add to Favorites Download image Share x Copy link
Relevant content

DeepSeek Splurges on Huawei Chips: 160,000 Ascend 950DT Chips Deployed in Inner Mongolia Supercomputing Cluster

Beating AI News Flash: Bloomberg cited people familiar with the matter that DeepSeek plans to deploy at least 160,000 Huawei Ascend 950DT chips at its AI data center under construction in Ulanqab, Inner Mongolia. If the plan is realized as scheduled, this will become one of the largest known Huawei AI chip clusters to date. DeepSeek also intends to purchase more chips, but Huawei’s production capacity cannot meet demand for now, with full delivery expected to take over a year. The 950DT chips will be used to run DeepSeek’s models; the company has no current plans to use them for model training. To date, DeepSeek still relies mainly on NVIDIA GPUs for core training. The 160,000 chips are only part of the overall project. DeepSeek previously planned to build a roughly 1GW-level AI data center in Ulanqab, and will rent additional computing power from other local operators.

3 minutes ago

ByteDance Secures $30 Billion Largest-Ever Loan: AI Cash Burn Hits Its Balance Sheet

Beating AI Express: ByteDance has secured a roughly $29.6 billion syndicated loan, setting a new company record and marking the second-largest dollar-denominated loan in Asia this year, trailing only SoftBank’s approximately $40 billion bridge loan. The firm had initially planned to borrow just $20 billion, but strong demand from participating banks pushed the total size up to nearly $30 billion. While the loan is not officially disclosed as being specifically for AI, with its stated purpose being general corporate use, ByteDance is sharply ramping up capital expenditure—focusing on expanding AI data centers and computing power infrastructure, while continuing to invest in models such as Doubao and Seedance, and expanding its enterprise-facing cloud business. Chinese tech giants’ AI financing activities are also continuing to scale up. Alibaba raised roughly $10.2 billion via a share placement last month, explicitly earmarking all proceeds for AI. ByteDance’s latest move, bringing its largest-ever loan to nearly $30 billion, signals that the AI race has evolved from competition in model capabilities to a battle over balance sheets.

3 minutes ago

Binance Alpha 1.0 lists CASHCAT, UP, and AI.

Binance Wallet announced in a post that Cash Cat (CASHCAT), UP (UP), and Artificial Inu (AI) have been added to Binance Alpha 1.0 on September 4, 2026.

3 minutes ago

Claude, despite being nearly 9 times more costly, still fell short: AIP2-1.0 leads the game creation agent segment in its first test.

Web3 project ClawQuest has launched AIP2-1.0, a game creation agent. Short for "AI Player Two", AIP2 operates on a model where players generate concepts and make choices, while the AI leverages tools to turn those ideas into functional, playable game content. AIP2-1.0 is now undergoing its first public trial in ClawQuest’s AI Agent title, Agent Fire. In testing, AIP2-1.0, GPT-5.6 Sol, and Claude Opus 5 were assigned the same 8 tank skill optimization tasks, each completing 300 final battles for a total of 2,400 matches per model. Results show AIP2-1.0 earned a composite score of 76.06, outperforming Claude Opus 5’s 73.96, and falling just 0.79 points short of GPT-5.6 Sol’s 76.85. For all 8 tasks, AIP2-1.0 incurred a model call cost of $20.69 — less than 12% of Claude Opus 5’s cost, and roughly 25% of GPT-5.6 Sol’s. Players can access AIP2-1.0 directly via the Agent Arena Bot on Telegram, with no need to deploy their own agent. ClawQuest aims to enable more players to participate in AI-powered game content creation through lower costs and a more streamlined entry point.

3 minutes ago

Adaption entrusts training set creation to AI: No raw data required, users can generate training sets simply by describing their goals.

Beating AI Express News: Adaption has launched Invent a Dataset, a training data generation tool. While standard synthetic data tools only handle sample generation, Invent a Dataset automates the full training set design process. Users simply describe what they want the model to learn; the tool independently designs the data structure and generates samples ready for direct model training. There’s no need to prepare raw data, design field formats, or pre-write detailed labeling rules upfront. For example, users can request the model to learn tasks like "summarizing business documents per company standards" or "routing customer service requests to relevant departments." Invent a Dataset identifies the required training data based on the user’s goal and generates corresponding samples in bulk. The generated data can also be directly input into Adaption’s AutoScientist, which will further adjust training data and parameters, iterating through training and evaluation until performance plateaus. This allows users to start from a single task description, generate a training set, and train a task-specific model end-to-end.

3 minutes ago

Unipcs showcases its latest achievements again: its account asset value has hit a new high of $20 million, with cumulative profits totaling $16.5 million.

Crypto KOL Unipcs (@theunipcs) posted that after his portfolio on the FOMO platform hit a $10 million all-time high yesterday, it has now reached a peak of $20 million, making him the first user on the platform to push his portfolio’s all-time peak past $20 million. He disclosed that his PNL over the past 24 hours stands at $4.7 million, $12.8 million over the past seven days, $14.8 million over the past 30 days, with cumulative PNL totaling $16.5 million. Key trades include returns of over 10,800% on PONS, 1168% on DELTA, 1015% on MICRODUCK, 350% on USELESS, 336% on MARSCOIN, and 119% on BASECAT.

3 minutes ago

Hot feeds

A trader profits $448K by monitoring #Binance's new listings!

2024.12.13 17:37:29

Last week, funds have flowed into #Bitcoin, #Ethereum, and #Hyperliquid.

2024.12.16 14:48:36

A $PEPE whale that had been dormant for 600 days transferred all 2.1T $PEPE($52M) to a new address.

2024.12.14 10:35:27

When Elon Musk tweeted about Moltbook, the meme coin MOLT experienced a short-term 30% price surge, hitting a new all-time high of $114 million.

2026.01.31 18:37:29

A smart #AI coin trader made $17.6M on $GOAT, $ai16z, $Fartcoin,$arc.

2025.01.05 16:05:18

A sniper earned 2,277 $ETH ($8.3M) trading $SHIRO within 18 hours!

2024.12.03 23:09:08

 More

Hot Articles

Follow us

X

Follow

Telegram

Join

Popular tokens

BitcoinEthereumHyperliquidSolanaTRONBNBTetherAaveXRPPepeFartcoinOndoJupiterUniswapBonkPendleEthenaArbitrumAvalancheLidoChainlinkPolygonDogecoinCardano