AI Daily · 2026-10-26

Anthropic filed an S-1 with the SEC for a proposed Nasdaq listing under the ticker ANT. Tencent Hunyuan released the Hy-SimulMT 30B A3B family of real-time speech-to-speech interpretation models, spanning GGUF and FP8 variants and emphasizing Chinese-English bidirectional translation, roughly 3-second streaming latency, and voice cloning. On the product side, Fanatics and OpenAI launched 'Shop the Look' across official league stores, with MLB tests showing 10% higher engagement than the previous version. LangChain also outlined five components of an internal 'company harness' for agents, covering identity, memory, sandboxed code, and scheduled triggers.

North America · First-hand

Anthropic

⭐⭐⭐ [Other] Anthropic files S-1 with the SEC for proposed initial public offering

Anthropic, PBC has filed a registration statement on Form S-1 with the U.S. Securities and Exchange Commission for a proposed initial public offering of its Class A common stock. The company intends to list on the Nasdaq Global Select Market under the symbol ANT. The number of shares to be offered and the price range have not yet been determined. The offering will be made only by means of a prospectus, copies of which will be available from Goldman Sachs & Co. LLC and Morgan Stanley & Co. LLC. The registration statement has been filed but is not yet effective, and no securities may be sold nor offers to buy accepted before that time.

OpenAI

⭐⭐ [Product Update] How Fanatics and OpenAI streamline the fan experience

Fanatics and OpenAI have collaborated on an OpenAI-powered version of Shop the Look, now live across many Fanatics sites including Fanatics.com and the official NFL, NBA, MLB and NHL online stores. When fans browse a team jersey, the feature suggests coordinated outfits on the product page using in-stock items matched to team and intended style, drawing on enriched product data plus team and inventory information, constrained by Fanatics merchandising rules, with looks refreshed daily. Fanatics uses ChatGPT Work and Codex across its Corporate, Commerce, Collectibles, and Betting and Gaming businesses; with Codex, the teams built custom tools to enrich product and catalog data, find compatible merchandise, evaluate recommendations and generate looks, testing the first 50 products in under a day and approving an MVP the same week after a couple days of iteration. After testing across roughly 30,000 MLB products, the new Shop the Look experience delivered 10% higher engagement than the previous one.

East Asia · First-hand

Tencent Hunyuan

⭐⭐⭐ [Model Release] tencent/Hy-SimulMT-30B-A3B-Interleave-GGUF

Tencent Hunyuan released Hy-SimulMT, a real-time simultaneous speech-to-speech translation model supporting bidirectional Chinese-English translation: it continuously consumes incoming audio and translates as the speaker talks, producing both translated text and speech, with real-time voice cloning that preserves the speaker's vocal characteristics. It is aimed at live settings such as meetings, speeches and interviews, seeking to keep translations accurate while reducing the disruption caused by waiting. On the open-source RealSI long-form audio benchmark it reached a StreamLAAL of about 3 seconds, the lowest among the compared systems, and its GEMBA score exceeded those of the compared open-source models and four of the five commercial APIs. On the short-form benchmarks FLEURS, CoVoST2, ACL6060 and RealSI-Sentence it scored 90.3, 75.5, 81.2 and 84.7 respectively. The GitHub repository provides the SimulInfer inference service and SimulTrain training tools; speech output and voice cloning require a configured TTS service, and the model can also be tried in the Hy Translate App. This release is the GGUF version of the model.

⭐⭐⭐ [Model Release] tencent/Hy-SimulMT-30B-A3B-Interleave-FP8

Tencent Hunyuan released Hy-SimulMT-30B-A3B-Interleave-FP8, a real-time simultaneous translation model supporting bidirectional Chinese-English speech-to-speech translation that takes in a continuous audio stream and outputs both translated text and speech as the speaker talks. It supports real-time voice cloning to preserve the speaker's vocal characteristics, and targets live settings such as meetings, speeches and interviews. On the open-source RealSI long-form audio benchmark it reached a StreamLAAL of about 3 seconds, the lowest among the compared systems, with a GEMBA score above the compared open-source models and four of the five commercial APIs. Short-form benchmark scores were FLEURS 90.3, CoVoST2 75.5, ACL6060 81.2 and RealSI-Sentence 84.7. The GitHub repository provides the SimulInfer inference service and SimulTrain training tools; speech output and voice cloning require a configured TTS service, and the model can also be tried in the Hy Translate App. It is released under apache-2.0.

⭐⭐⭐ [Model Release] tencent/Hy-SimulMT-30B-A3B-Interleave

Tencent Hunyuan released Hy-SimulMT, a 30B (A3B sparse) Apache-2.0 model for real-time simultaneous translation that supports bidirectional Chinese-English speech-to-speech translation. It continuously receives incoming audio and translates as the speaker talks, producing both translated text and speech, and uses real-time voice cloning to preserve the speaker's vocal characteristics; it targets live settings such as meetings, speeches and interviews. On the open-source RealSI long-form audio benchmark, Hy-SimulMT reached a StreamLAAL of about 3 seconds, the lowest among the compared systems, and its GEMBA score exceeded the compared open-source models and four of the five commercial APIs. The GitHub repository provides the SimulInfer inference service and SimulTrain training tools, speech output and voice cloning require a configured TTS service, and the model can also be tried in the Hy Translate App.

⭐⭐⭐ [Model Release] tencent/Hy-SimulMT-30B-A3B-GGUF

Tencent Hunyuan released Hy-SimulMT, a speech-to-speech model for bidirectional Chinese-English real-time simultaneous translation that continuously takes in audio and translates as the speaker talks, outputting both translated text and speech, with real-time voice cloning that preserves the speaker's vocal characteristics. On the RealSI long-form audio benchmark it reached a StreamLAAL of about 3 seconds, the lowest among the compared systems, and its GEMBA score exceeded those of the compared open-source models and four of the five commercial APIs. On four short-form audio benchmarks it scored 90.3 on FLEURS, 75.5 on CoVoST2, 81.2 on ACL6060 and 84.7 on RealSI-Sentence. The GitHub repository provides the SimulInfer inference service and SimulTrain training tools; speech output and voice cloning require a configured TTS service, and the model can also be tried in the Hy Translate App.

⭐⭐⭐ [Model Release] tencent/Hy-SimulMT-30B-A3B

Tencent Hunyuan has released and open-sourced Hy-SimulMT-30B-A3B (Apache-2.0), a real-time simultaneous translation model supporting bidirectional Chinese–English speech-to-speech translation. It continuously consumes incoming audio, translating as the speaker talks and producing both translated text and speech, with real-time voice cloning that preserves the speaker's vocal characteristics; it targets live settings such as meetings, speeches and interviews. On the open-source RealSI long-form audio benchmark, Hy-SimulMT reached a StreamLAAL of roughly 3 seconds, the lowest among the compared systems, and its GEMBA score exceeded those of the compared open-source models and four of the five commercial APIs. The post also reports comparison scores on four public short-form audio benchmarks — FLEURS, CoVoST2, ACL6060 and RealSI-Sentence (e.g., 90.3, 75.5, 81.2 and 84.7). The GitHub repository provides the SimulInfer inference service and SimulTrain training tools, speech output and voice cloning require a configured TTS service, and the model can also be tried in the Hy Translate App.

⭐⭐ [Model Release] tencent/Hy-SimulMT-30B-A3B-FP8

Tencent Hunyuan released Hy-SimulMT, a real-time simultaneous translation model (30B, FP8, Apache-2.0) supporting bidirectional Chinese–English speech-to-speech translation: it takes in a continuous audio stream and produces translated text and speech while the speaker talks, with real-time voice cloning that preserves the speaker's vocal characteristics. On the open-source RealSI long-form audio benchmark it reaches a StreamLAAL of about 3 seconds, the lowest among the compared systems, and its GEMBA quality score exceeds those of the compared open-source models and four of the five commercial APIs. On four short-form audio benchmarks it scores 90.3 (FLEURS), 75.5 (CoVoST2), 81.2 (ACL6060) and 84.7 (RealSI-Sentence), versus 88.4, 82.1, 80.5 and 84.4 for Qwen3.8-LiveTranslate. The GitHub repository provides the SimulInfer inference service and SimulTrain training tools, speech output and voice cloning require a configured TTS service, and the model can also be tried in the Hy Translate App.

Ecosystem & Beyond (Products / Agents / Tools / Opinions)

Other

⭐⭐ [Other] The five components of building a "company harness"

LangChain describes its experience building internal agents for marketing, sales and engineering, distilling five shared primitives for a company harness: the agent lives where people already work (Slack for them), it knows who is asking and acts on behalf of that person for some services while using its own identity for others, it has memory scoped globally and/or per user, code runs in a sandbox and can call services that require auth, and a large share of runs are scheduled rather than explicitly invoked. Three internal agents are profiled: a GTM Agent that monitors and answers email, responds to qualified leads, sends daily briefs and runs outbound motions; OpenSWE, an engineering agent that drafts and reviews PRs; and Content Studio, a marketing agent that drafts blogs, images, slide decks and videos in company style. The authors walk through problems one by one: for UX they added a way for agents to break out into their own Slack thread and built a custom UI for backlog-heavy workflows; for identity they contrast on-behalf-of actions with fixed credentials, detailing per-agent choices such as GTM Agent's email tools being available only in DMs and Content Studio restricting certain Twitter operations to whitelisted users. The post is truncated while discussing memory.