Tag
Anthropic 发布 Claude Platform 的 Message Threads 公测功能，通过仅发送增量消息而非重发完整对话历史，将长会话的时间到首 token（TTFT）中位数降低约 35%。
Anthropic is open-betaling "message threads" for the Claude Messages API: the server stores conversation history, so each turn developers only send new messages plus the previous reply ID. This significantly cuts latency on long conversations (median TTFT for 500K–1M token requests drops by ~35%), with pricing and rate limits unchanged.
Hume AI 发布了 Hume Audio Decision API，开发者只需一次调用即可让 AI 代理理解音频内容（如说话人数、笑声、情绪等），取代过去需要 10 多个模型的流程，该 API 基于 1000 万+ 人类判断训练，支持实时与长音频分析。
Hume AI introduced the Hume Audio Decision API, a single-call model that takes audio plus a list of questions and returns answers with confidence scores, detecting tone, hesitation, laughter and speaker cues without needing a transcript.
Hume AI announced the Hume Audio Decision API, an audio-native model that answers questions about tone, emotion, language, and speaker count directly from audio without needing a transcript, trained on 10M+ human judgments and 100M hours of audio.
A free, daily-updated AI pricing and usage tracking API (modelmonitor.dev) that provides pricing across providers and models, intended for integration into text editors to help developers choose the right model while coding.
Show HN: An updated website and API (syscall.sh) that aggregates Linux system call tables for 27 architectures, covering Linux 3.3 to 7.3, with numbers, arguments, registers, calling conventions, and machine-readable JSON exports for each architecture.
humans& releases &Tutor, an AI model trained to teach by training against Persimmon as a simulated learner, achieving state-of-the-art results on the TutorMoments benchmark and outperforming frontier models like Claude Sonnet 5.5 and GPT-6 Astra under plain prompting.
The Claude API has increased the per-request size limit for Messages and Token Counting from 32MB to 128MB, roughly a fourfold increase, to support tasks such as long conversations containing many images and large PDFs.
River API now supports training LLMs with arbitrary custom loss functions written in PyTorch, which are traced locally, sent as a graph, and executed at scale on River's remote GPU cluster with parallelization and memory-efficient streaming.
Santiago (@svpino) complains about the overwhelming amount of AI-generated replies on X, and the post asks whether removing the ability to reply via the API would improve the situation.
Mischief 1 is introduced as a commercial image-to-video model built on four years of research into biological neural networks on electrodes, priced from $0.05 per second and claimed to be at least 5x faster than its base model. It will be available via fal, OpenRouter, and soon the AWS Marketplace, positioning 'biocompute' as a real commercial technology.
PokeBot introduces RoboAPI, powered by its Poke-Aurora world-action model, allowing owners of 5 supported robot arms to send natural-language commands via a simple API at 30Hz — no task-specific data collection or training required.
The author of SocialCrawl introduces 'Discover', a cross-platform social listening API endpoint for AI agent builders that searches multiple platforms, scores posts against natural-language questions, dedupes, and returns a unified result list with spend controls. Feedback and free API credits are being offered for testers.
Learn how to build a live stock heatmap app using Massive for real-time financial data and Claude Opus 5.5 to generate the application code, with an easy path to extend it further.
Nace 发布 Drex Max 1.1，这是一个在分布外泛化上显著提升的多模态决策模型，拥有 262K 上下文窗口，并在 Decision Index 上取得 70.8 的成绩。
A sponsored post comparing REST, GraphQL, gRPC, and WebSockets, then introducing Deepgram Flux TTS — a conversation-aware text-to-speech API for real-time voice agents with built-in interruption handling, sub-80ms latency, and VPC/on-prem deployment.
Parallel Web Systems launched its Decisions API, a web-grounded decision engine that returns confidence scores (e.g., 98% or 55%) for yes/no answers, letting apps auto-trust confident responses and route uncertain ones to humans at $0.001 per request.
Chandra OCR 2.6 发布，宣称在 API 上比前沿模型更快、更便宜且更准确，超越 Gemini Flash、Opus 和 Sol，成本降低 5-15 倍，速度提升 2-10 倍。
Modal 发布 Spindle，一个开源、与 Tinker 兼容的训练服务器，可在 Modal 工作区中运行，让现有 Tinker 脚本无需改动即可迁移，并以 GPU 时间计费取代按 token 计费。