Two internal OpenAI coding agents switched off safety monitoring on their own, even after weighing whether to keep the safeguards in place. When asked about it, GPT-6 Astra denied removing the monitoring. OpenAI calls this a recurring pattern and doesn't yet have a satisfying explanation for the behavior. The article OpenAI can't yet explain why its coding agents switch off misalignment monitoring appeared first on The Decoder.
Anthropic's Claude independently designed protein binders targeting deadly scorpion neurotoxins. Out of more than 318,000 designs, 211 were tested in the lab. Seventeen met the criteria, and the two strongest bind extremely tightly to Cn2, a toxin from a Mexican scorpion. Whether they actually work as antivenom remains an open question. The article Claude Science tackles deadly scorpion stings by designing protein binders from scratch appeared first on The Decoder.
OpenAI is acquiring startup Orchestra Bio and plans to merge it with its life sciences AI platform Rosalind. The goal is to build AI agents that help pharma teams coordinate drug development, from tracking trials to spotting problems. Developing a new drug in the US takes ten to 15 years on average. OpenAI already works with Novo Nordisk and Moderna on AI-driven pharma research. The article OpenAI acquires Orchestra Bio to build AI agents that speed up drug development appeared first on The Decoder.
Mistral AI just released Voxtral 3. Mistral Voxtral 3 is a speech model family paired with a voice agent SDK. It ships 3 transcription models, 1 text-to-speech model and an SDK built on LiveKit. One place for listening, reasoning, tool calls and speaking. All 4 models are in public preview on the Mistral API. TL;DR Size: Voxtral Mini Transcribe 3 has 4B parameters. Small Transcribe 3 size: not disclosed. Both batch models list a 32k context. Runs on: Mistral API (EU, US and global endpoints) or your own infrastructure via Mistral sales. GPU and VRAM needs: not disclosed. Mistral says Mini suits resource-constrained environments. Performance: Top 3 on independent speech-to-text rankings, strongest on diarization and multilingual batch transcription. Best: 2.1% AA-WER for Small Transcribe 3, #3 on Artificial Analysis. Worst: Realtime at 240 ms: 7.4% WER on FLEURS 35, vs 4.2% for Gemini 3.5 Realtime. What is in the Voxtral 3 family? Mistral lists 4 models and 1 SDK: Voxtral Small Transcribe 3: flagship batch transcription and diarization. Docs also list translation, speech Q&A, summarization and function calling. Voxtral Mini Transcribe 3: a 4B model for cheaper, faster batch jobs. Voxtral Mini Transcribe Realtime 3: streaming ASR over WebSocket with end-of-turn detection. Voxtral Speech 3: text-to-speech, also called Voxtral Mini TTS 3 in the post. All transcription models support diarization and context biasing. Context biasing takes key terms or free text to improve recall. Batch transcription covers 50 languages, including 24 official EU languages. How does the Voxtral voice agent SDK work? The SDK is built on LiveKit and runs 3 stages as 1 streaming pipeline. Realtime Transcribe 3 listens and detects when the caller stops speaking. A general LLM, such as Mistral Small 4, reasons and calls tools. Voxtral Speech 3 replies in the tone of the conversation so far. Each turn completes in under 1 second. The agent stops talking when the caller cuts in. It handles phone calls over SIP and browser or app sessions over WebRTC. Because each stage is a separate model, teams can swap the LLM, voice or language. (function(){var f=document.getElementById('mtp-vx3-frame');window.addEventListener('message',function(e){if(e.data&&e.data.mtpEmbed==='mtp-vx3'&&e.source===f.contentWindow){f.style.height=e.data.h+'px';}});})(); How accurate is Voxtral Transcribe 3? These numbers are self-reported by Mistral: FLEURS 45 (WER, forced / inferred language): Small 4.1% / 4.0%. MAI-Transcribe-2 ties at 4.1% / 4.0%. Mini 4.6% / 4.5%. Whisper large-v3 16.0% / 14.9%. Diarization (cpWER, 5 public sets): Small 18.4% and Mini 18.6%. Pyannote scores 19.0% and MAI-Transcribe-2 20.3%. Keyphrase recall (Earnings 22): Small 94.6% with context, 81.4% without. Scribe v2 posts 94.3% and 84.0%. On Artificial Analysis, Small scores 2.1% AA-WER and Mini 2.2%. Only Fun-Realtime-ASR-preview (1.7%) and MAI-Transcribe-2 (2.0%) rank higher. Speed is the gap: Small’s speed factor is 65.8x, against 307.5x for MAI-Transcribe-2. How do realtime transcription and TTS compare? On FLEURS 35, the realtime model trades latency for accuracy. It scores 5.7% WER at 2,400 ms and 6.4% at 480 ms. At 240 ms, WER rises to 7.4%. Gemini 3.5 Realtime Transcribe scores 4.2% and MAI-Transcribe-2 Streaming 5.8%. GPT Live Transcribe (8.4%) and Scribe v2 Realtime (8.7%) trail every Voxtral setting. Voxtral Speech 3 costs $13 per 1M characters. Mistral says that undercuts OpenAI TTS-1, Deepgram, ElevenLabs Flash and Gemini Flash. Explicit controls cover emotion, pacing and emphasis. The model page lists ~90 ms time-to-first-audio and 34 emotion tags. Pro Voice adapts a speaker’s style from 30 minutes of audio. Artificial Analysis places it in the top 12 of both TTS arenas. Voxtral 3 vs MAI-Transcribe-2, Scribe v2 and Gemini 3.5 Transcribe How much does Voxtral 3 cost, and where can you run it? Per Mistral’s price table: Voxtral Mini Transcribe 3 (batch): $0.0035/min Voxtral Small Transcribe 3: $0.004/min Voxtral Mini Transcribe Realtime 3: $0.008/min Voxtral Speech 3 (batch and realtime): $13 per 1M characters Voice agent API: $0.04/min with Mistral Small 4, $0.08/min with any other reasoning model The models run on Mistral’s European, US and global endpoints. OpenRouter also lists all 3 new models with EU-hosted, zero-data-retention endpoints. Key Takeaways Voxtral 3 bundles 3 ASR models, 1 TTS model and a LiveKit voice agent SDK. Small Transcribe 3 hits 2.1% AA-WER, #3 on Artificial Analysis. Best-in-chart diarization: 18.4% cpWER, ahead of Pyannote and MAI-Transcribe-2. Batch transcription costs $0.0035 to $0.004 per minute; TTS costs $13 per 1M characters. Realtime accuracy trails Gemini 3.5 Realtime, and no open weights were announced. Check out the Voxtral 3 announcement, the audio docs and the model pages. All credit goes to the researcher of this project. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well. The post Mistral AI Releases Voxtral 3: Speech-to-Text, TTS and a Voice Agent SDK appeared first on MarkTechPost.
Elevenlabs' new Eleven Realtime speech-to-speech models combine speed and accuracy. The v1 Max variant scores 98.8 percent on the Big Bench Audio benchmark with a 0.57-second response time. Elevenlabs says no model on the Artificial Analysis leaderboard is both faster and more accurate. The article Elevenlabs releases two new speech-to-speech models designed for real-time AI conversations appeared first on The Decoder.
Deepseek is open-sourcing DeepInfer, the inference framework behind its fast V4.1-Flash model. The vLLM-based system cuts compute costs and energy use when running AI models. The release adds more pricing pressure on Western providers like OpenAI and Anthropic. The article Deepseek open-sources the inference engine that powers its fastest and cheapest AI model appeared first on The Decoder.
Mistral AI launched Voxtral 3, a new family of language models built for real-time voice agents. The system combines speech-to-text, text-to-speech, and a voice agent SDK, supports more than 50 languages, and lets agents tap into external tools during live conversations. It can run on your own infrastructure or through EU-hosted APIs to keep data under your control. The article Mistral AI launches Voxtral 3, a full voice agent stack that listens, thinks, and talks in real time appeared first on The Decoder.
OpenAI CEO Sam Altman laid out three paths to safe AI in an internal presentation: an international watchdog modeled after the IAEA, a merger of the major labs into one entity, or a single company so powerful it could shut down everyone else's compute. He called the last option "terrible." The article Sam Altman reportedly pitched three endgames for AI safety, one involving a single firm melting everyone else's GPUs appeared first on The Decoder.
OpenAI improved its latest models for low-resource programming languages like COBOL, Fortran, and Visual Basic. On the new MultiLangBench, GPT-6 Astra hits an 87.4 percent test pass rate across 20 languages, up from about 38 percent for GPT-5.6 Terra. For COBOL migration specifically, GPT-6.1 Sol reaches up to 97.9 percent accuracy, a result that could matter for modernizing critical financial infrastructure. The article OpenAI says GPT-6 is much better with legacy programming languages such as COBOL and Fortran appeared first on The Decoder.
