4OpenAI Release Notes·
OpenAI extends default prompt cache retention to 24 hours for non-ZDR organizations
OpenAI updated its API default for prompt_cache_retention from in_memory to 24h for organizations without Zero Data Retention (ZDR) enabled. This change makes extended prompt caching the default behavior, reducing redundant token processing for repeated prompt prefixes. The update has practical cost and latency implications for API users who rely on prompt caching.
Related guides (3)
Related events (8)
Prompt Caching in the API
OpenAI is introducing automatic prompt caching for API users, providing discounts on input tokens that the model has recently processed. The feature reduces costs for repeated or overlapping prompt prefixes without requiring explicit developer configuration. This follows Anthropic's similar caching feature and reflects broader industry movement toward inference cost optimization.
Anthropic Publishes Quantitative Case Study on Prompt Engineering for Long-Context Recall
Anthropic shares a quantitative case study evaluating prompting techniques to improve Claude's recall over 75,000–90,000 token contexts. Two techniques are tested: extracting reference quotes before answering, and providing few-shot examples of correctly answered questions. The study uses Claude Instant 1.2 on a government document dataset constructed via a 'randomized collage' method, with multiple-choice Q&A pairs generated by Claude itself. Results show measurable recall improvements over a baseline prompt, with methodology and notebooks shared publicly.
OpenAI rolls out improved persistent memory for ChatGPT Enterprise and Edu
OpenAI is deploying enhanced memory capabilities to ChatGPT Enterprise and Edu tiers, enabling the system to automatically surface and use relevant context from past conversations rather than relying solely on manually saved details. Users gain a memory summary view, source attribution for personalized responses, and controls to correct, delete, or disable memory. Enterprise rollout begins with a two-week admin-gated early access period before defaulting on. The update is included at no additional cost.
Understanding prompt injections: a frontier security challenge
OpenAI has published a blog post addressing prompt injection attacks as a key security challenge for AI systems. The post covers how these attacks work and outlines OpenAI's multi-pronged approach including research, model training improvements, and safeguard development. This signals OpenAI's formal positioning on agentic security threats as their models are increasingly deployed in tool-using and autonomous contexts.
OpenAI rolls out improved automatic memory for ChatGPT Business
OpenAI is expanding ChatGPT Business with improved memory that automatically uses context from past conversations rather than relying solely on manually saved details. Users gain a memory summary view, source attribution for personalized responses, and controls to correct, delete, or disable memory references. The update is available at no additional cost and does not affect Codex or project-scoped memory.
OpenAI deprecates reusable prompts, Evals platform, and Agent Builder
OpenAI announced the deprecation of three platform features: reusable prompt objects, the Evals platform, and Agent Builder, with shutdown timelines and migration guidance available on their deprecations page. This signals a consolidation or strategic shift in OpenAI's developer tooling and evaluation infrastructure. The removal of the Evals platform is particularly notable for practitioners who relied on it for structured model evaluation workflows.
OpenAI adds Admin API capabilities for spend alerts, model allowlists, and billing
OpenAI released new Admin API features enabling enterprise administrators to manage spend alerts, model allowlists, data retention settings, and hosted tool permissions. The update also adds querying of granular billing line items. These additions expand programmatic control for organizations deploying OpenAI at scale.
Claude Code sends 33k tokens before reading the prompt vs. OpenCode's 7k
A blog post with significant HN traction (329 points, 185 comments) compares the token overhead of Claude Code and OpenCode before any user prompt is processed, finding Claude Code consumes ~33k tokens versus OpenCode's ~7k. The analysis highlights a substantial efficiency gap in system prompt and context initialization between the two coding agents. This matters for inference cost and latency in agentic coding workflows.