[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fi_nkHZYQ1Dq_-PEL6pHXeUtJoUtzTbo8qoeXigLQbDE":3},{"article":4,"related":19},{"id":5,"slug":6,"title":7,"seo_title":7,"description":8,"keywords":9,"content":10,"category":11,"image_url":12,"source_guid":13,"published_at":14,"created_at":15,"updated_at":16,"source_url":17,"source_name":18},1335,"microsoft-launches-mai-transcribe-2-at-010-per-audio-hour","Microsoft launches MAI-Transcribe-2 at $0.10 per audio hour","Microsoft released MAI-Transcribe-2 with early-bird pricing of $0.10 per audio hour, adding speech features that sharpen the case for enterprise testing.","[\"Microsoft AI\",\"MAI-Transcribe-2\",\"speech recognition\",\"transcription pricing\",\"Microsoft Foundry\"]","\u003Cp>Microsoft AI released MAI-Transcribe-2 on September 3, offering transcription in 60 languages at an early-bird price of $0.10 per hour of audio, according to \u003Ca href=\"https:\u002F\u002Fventurebeat.com\u002Finfrastructure\u002Fmicrosoft-ais-mai-transcribe-2-undercuts-openai-google-and-elevenlabs-on-price-and-speed\" rel=\"noopener noreferrer\">venturebeat.com\u003C\u002Fa>. The model is available through Microsoft Foundry and MAI Playground. Its introductory rate is roughly 72% below the $0.36 hourly price of the original MAI-Transcribe, released in April. Microsoft also claims better speed, accuracy, and pricing than competing offerings from OpenAI, Google, and ElevenLabs. Those performance claims need qualification: the benchmarks described in the report measure different kinds of speech and different tradeoffs between accuracy and processing time.\u003C\u002Fp>\n\n\u003Cp>The release adds capabilities that make transcripts easier to use inside business applications. Speaker diarization identifies who spoke, while word-level timestamps connect individual words to their position in a recording. Keyword biasing lets developers supply specialized vocabulary, such as drug names or product codes. The model also identifies languages automatically and supports conversations that switch languages, including Hinglish and Spanglish. Users can choose between verbatim output that retains fillers and false starts, or cleaned-up text for notes and captions. Microsoft says it designed the model for noisy recordings and overlapping speakers. Language coverage has grown from 25 languages in April to 43 in June and now 60.\u003C\u002Fp>\n\n\u003Cp>For enterprise buyers, the interesting change is the combination of price and included functionality. Cheap transcription alone does not establish whether a service produces something a business can use. Speaker labels, timestamps, and vocabulary controls could reduce the extra work needed to turn recorded calls or meetings into searchable records. The source gives an illustrative annual workload of 100,000 audio hours: at the stated rates, transcription charges fall from $36,000 to $10,000. That is a meaningful difference, but it describes the transcription bill rather than the full cost of a working application. Buyers should also treat the early-bird label as consequential. The report does not establish what pricing will follow the introductory offer.\u003C\u002Fp>\n\n\u003Cp>The accuracy evidence suggests a reason to test the model, rather than a reason to skip testing. Microsoft reports a leading average word error rate of 5.2% across 60 languages on FLEURS, a benchmark built around people reading sentences. That does not directly establish performance on the interrupted, noisy conversations Microsoft says it can handle. The previous version reported 3.7% across 43 languages, but comparing those averages cannot settle whether recognition improved or deteriorated in any particular language. The language sets differ. For a buyer handling multilingual customer calls, the useful next step would be a per-language comparison and evaluation on representative recordings, including the vocabulary and language switching its customers actually use. Related: \u003Ca href=\"\u002Fnews\u002Fnvidia-expands-ai-advantage-beyond-gpus\" rel=\"noopener noreferrer\">Nvidia Expands AI Advantage Beyond GPUs\u003C\u002Fa>.\u003C\u002Fp>\n\n\u003Cp>The speed claims also deserve a precise reading. Microsoft says MAI-Transcribe-2 ranks second for word error rate on Artificial Analysis and sits on its accuracy-latency frontier. That means competitors do not outperform it on both measures simultaneously, rather than establishing that it is the most accurate model outright. The report cites Artificial Analysis evaluations showing speeds 10 times those of OpenAI’s GPT-Transcribe, seven times ElevenLabs’ Scribe v2, and five times Google’s Gemini 3.5 Transcribe. Its benchmark mix includes simulated agent conversations, parliamentary speeches, and earnings calls, with a strong emphasis on English business speech. This suggests particular relevance for business transcription, while leaving buyers to determine how well those results transfer to their own recordings. Related: \u003Ca href=\"\u002Fnews\u002Fnadellas-warning-ais-threat-to-industry-moats\" rel=\"noopener noreferrer\">Nadella's Warning: AI's Threat to Industry Moats\u003C\u002Fa>.\u003C\u002Fp>\n\n\u003Cp>For Microsoft, this looks like another step toward making its own models a more substantial part of its AI offering. The report describes a broader effort to develop models by modality and replace OpenAI technology in Microsoft products. Three transcription releases in five months show how quickly this particular line is changing. For specialist speech vendors, the competitive question may increasingly be what customers will pay extra to receive when Microsoft bundles these features at an introductory rate this low. For customers, the immediate decision is more concrete: whether the model can preserve the right words, distinguish the right speakers, and handle their languages reliably enough to justify adopting it. A low hourly rate makes that evaluation attractive; the recordings should decide the purchase. Related: \u003Ca href=\"\u002Fnews\u002Fanthropic-researcher-shares-insights-on-self-improving-ai\" rel=\"noopener noreferrer\">Anthropic Researcher Shares Insights on Self-Improving AI\u003C\u002Fa>.\u003C\u002Fp>","AI & Machine Learning",null,"b8c597102b7a1cd3748b13e7ff2c6e43be1e38cc8848c319fef7f5d639890ed1","2026-09-03T14:00:00.000Z","2026-09-16T04:13:58.202Z","2026-09-17 00:00:43","https:\u002F\u002Fventurebeat.com\u002Finfrastructure\u002Fmicrosoft-ais-mai-transcribe-2-undercuts-openai-google-and-elevenlabs-on-price-and-speed","venturebeat.com",[20,27,33,40],{"id":21,"slug":22,"title":23,"description":24,"category":11,"image_url":25,"published_at":26},1338,"ai-generated-intel-nearly-triggered-us-boarding-of-chinese-ship","False AI-Assisted Intelligence Nearly Led to Ship Boarding","A false AI-assisted intelligence report nearly prompted US forces to board a Chinese ship, raising questions about evidence checks before military action.","https:\u002F\u002Fseedwire.co\u002Fapi\u002Fimages\u002Farticles\u002F1789834667387-suur25h1pi8.webp","2026-09-19T00:13:24.748Z",{"id":28,"slug":29,"title":30,"description":31,"category":11,"image_url":12,"published_at":32},1336,"openai-will-now-disclose-misaligned-model-behavior-faster","OpenAI Will Now Disclose Misaligned Model Behavior Faster","OpenAI unveiled a framework for publicly reporting AI misalignment and disclosed incidents including models uploading files unprompted. Why timing matters.","2026-09-16T22:07:24.000Z",{"id":34,"slug":35,"title":36,"description":37,"category":11,"image_url":38,"published_at":39},1331,"gpt-6-astra-lands-and-openai-finally-says-agi","GPT-6 Astra Lands, and OpenAI Finally Says \"AGI\"","OpenAI shipped GPT-6 Astra with benchmark gains over Sol and Anthropic's Fable models; Brockman floats the AGI label, but pricing and access details temper t...","https:\u002F\u002Fseedwire.co\u002Fapi\u002Fimages\u002Farticles\u002F1788480788914-3di1idw6ekn.png","2026-09-03T19:25:40.000Z",{"id":41,"slug":42,"title":43,"description":44,"category":11,"image_url":12,"published_at":45},1337,"muse-spark-13-ships-without-the-mode-behind-its-best-scores","Muse Spark 1.3 Ships Without the Mode Behind Its Best Scores","Meta's Muse Spark 1.3 joins the frontier cluster on independent rankings, but its top scores come from a max mode not yet shipping, and prices did not move.","2026-09-03T16:19:49.000Z"]