Source details
- Original source
- Towards AI
- Published
- 2026-09-25
- Primary topic
- Foundation Models
Why it matters
Model launches, benchmark jumps, API upgrades, context window changes, and frontier LLM competition. Use the original source for the full report, then use the directory shortcuts below to compare the products and workflows the story points toward.
What happened
Author(s): Chew Loong Nian – AI ENGINEER Originally published on Towards AI. Anthropic’s SDK 0.128.0, released with Claude Opus 5.5, can hand the model a new tool mid-run without touching tools[]. It needs one beta flag the runner won’t add for you. If your Claude agent picks up a new tool halfway through a long conversation, the way you add it decides how much of the next request can still come from the prompt cache: 98.7% of it with the new runner.addTools(), none if you edit tools[]. After introducing the problem of tool changes during long runs, the article explains what changed in the Anthropic SDK (TypeScript 0.128.0) and why editing tools[] is expensive: tool definitions are at the front of the prompt-cached prefix, so changing them forces a full cache miss. It then shows how addTools() avoids that by leaving tools[] intact and instead appending a system message containing a tool_addition block, so only the appended part is processed as new input. The author provides and walks through a reproducible Node script that measures request JSON reuse, demonstrating up to 98.7% shared prefix with addTools() versus near-zero reuse when updating tools[], and shows how the gap grows with conversation length. Key caveats include needing the inline-tools-2026-09-15 beta header/param (not auto-added by the runner), model limitations, and behavioral details like one special case that can still cause a full miss, plus how pause turns delay tool change propagation while both addition/removal are immediate on the agent side. The piece ends with a practical verdict: keep your initial tools[] unchanged and use addTools(), but add the beta flag explicitly for supported models. Read the full blog for free on Medium. Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor. Published via Towards AI
What to do next
Compare the hosted model pages first, then check the related tools and buyer guides before changing workflow standards.
Author(s): Chew Loong Nian – AI ENGINEER Originally published on Towards AI. Anthropic’s SDK 0.128.0, released with Claude Opus 5.5, can hand the model a new tool mid-run without touching tools[]. It needs one beta flag the runner won’t add for you. If your Claude agent picks up a new tool halfway through a long conversation, the way you add it decides how much of the next request can still come from the prompt cache: 98.7% of it with the new runner.addTools(), none if you edit tools[]. After introducing the problem of tool changes during long runs, the article explains what changed in the Anthropic SDK (TypeScript 0.128.0) and why editing tools[] is expensive: tool definitions are at the front of the prompt-cached prefix, so changing them forces a full cache miss. It then shows how addTools() avoids that by leaving tools[] intact and instead appending a system message containing a tool_addition block, so only the appended part is processed as new input. The author provides and walks through a reproducible Node script that measures request JSON reuse, demonstrating up to 98.7% shared prefix with addTools() versus near-zero reuse when updating tools[], and shows how the gap grows with conversation length. Key caveats include needing the inline-tools-2026-09-15 beta header/param (not auto-added by the runner), model limitations, and behavioral details like one special case that can still cause a full miss, plus how pause turns delay tool change propagation while both addition/removal are immediate on the agent side. The piece ends with a practical verdict: keep your initial tools[] unchanged and use addTools(), but add the beta flag explicitly for supported models. Read the full blog for free on Medium. Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor. Published via Towards AI
This AimostAll brief summarizes the linked source so readers can scan AI developments quickly and jump to the original reporting when needed.
