975 summarised stories in Tools & Coding, each linking back to the original source. Browse all topics →
OpenAI 1 year ago 35
OpenAI released a Realtime API that enables developers to build applications with fast speech-to-speech capabilities. The API supports bidirectional audio streaming with response times measured in hundreds of milliseconds. This allows developers to create interactive voice applications with lower latency than previous API options.
OpenAI 1 year ago 4
OpenAI's fine-tuning API now supports GPT-4o with both images and text inputs, allowing developers to customize the model's vision capabilities. The update enables model training on multimodal datasets containing 1 to 500 training examples per task. This expands the fine-tuning options beyond text-only models to include vision-based applications.
OpenAI 1 year ago 35
Anthropic added prompt caching to its API, which reduces costs for repeated input sections by storing them in the model's cache. Cached tokens cost 90% less than standard tokens after a 1,024-token minimum, and cached content remains available for five minutes. Users can lower expenses for applications that process the same documents, codebases, or instructions repeatedly.
OpenAI 1 year ago 28
OpenAI now allows users to fine-tune smaller, cheaper models using outputs generated by larger frontier models through its API. The process runs entirely on OpenAI's platform, eliminating the need for external infrastructure. This enables developers to create cost-efficient models that replicate the capabilities of expensive large models without paying for repeated calls to those frontier models.
The daily briefing
TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.
TLDRocket
8 August 2026
Mistral AI's release of Shieldstral 1.0 3B marks a practical inflection point in content moderation: safety classifiers are shedding their fixed rulebooks. The 3B model takes policies as plain-language prompts at inference time—meaning a single deployment can enforce different standards for different customers without retraining—while matching the performance of a 20B baseline on both text and multimodal content. It runs on a single 16GB GPU and ships under Apache 2.0, making it available for local deployment across vLLM and llama.cpp. For teams managing multiple content policies or operating across jurisdictions, this matters because moderation no longer requires separate model forks or constant retraining cycles.
Read the full briefing →