Agent APIs reduce session overhead without making context free
Claude message threads reduce repeated context uploads without reducing token accounting, while OpenAI’s Agents API adds mutable tools, background compaction and separately billed automatic review.
Anthropic’s November 2 message-threads beta lets an agent send the full conversation once and then only new messages plus the previous response identifier.
Threads remain encrypted in memory and expire after five minutes without a request, but deletion can take up to one hour. They are not eligible for zero data retention. Claude still processes the full conversation on each turn, so ordinary token usage and rates remain; a smaller network request does not imply a smaller model bill. Anthropic reports about 35% lower median time to first token for Claude Code requests carrying 500,000–1 million tokens, a specific vendor observation rather than a general latency promise. [1][2]
OpenAI’s October 30 Agents API update addresses a different layer: an existing session can replace its tool list, and context compaction runs in the background by default. Function tools and MCP changes take effect on the next model call; built-in tool changes take effect by the next turn. Background compaction may increase token usage. [3][4]
Sources & details 5 sources
Discuss in your voice app
Take the reporting, visual explanations and original sources into a conversation. Copy the brief, paste it into your preferred assistant, then start its voice mode.
Paste the brief into your voice app to discuss this reporting.Reporting sources
- Anthropic — Message threads
November 2 public beta, five-minute expiry, unchanged token accounting and vendor latency observation.
- Claude Platform — Message threads
Request lifecycle and retained/fixed fields; same provider, not an independent latency replication.
- OpenAI — API changelog
October 30 mutable tools, asynchronous compaction and automatic-review updates.
- OpenAI — Agent configuration
Tool replacement and compaction configuration.
- OpenAI — Automatic review
Eligibility, supported review requests, limits and billing.
Images & figures
Claude threads upload the conversation once and send only additions on later turns. The model still receives the reconstructed full context and token pricing is unchanged; the block counts are illustrative.
Euangelion · Source · Usage terms