diff --git a/CHANGELOG.md b/CHANGELOG.md
index 1c57101..412fc25 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -4,6 +4,12 @@ This project follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/) fo
 
 ## [Unreleased]
 
+## [0.61.1] - 2026-09-10
+
+### Fixed
+
+- **`dispatch` never checked the TPM ceiling it already knew.** `pick` has refused to *seat* an oversized prompt since `token_limits` landed, but `dispatch` — reachable directly by any caller naming an agent id, skipping `pick` entirely — sent the request anyway. Two live 413s on `groq-gpt-oss-120b` (2026-09-08/09) requested 17124 and 22583 tokens against an observed ceiling of 8000 that had not moved in days: the ceiling was known, nothing had consulted it. `runAny` now runs the same `effectiveTokenCeiling` check right after assembling the full prompt and refuses before dispatching, with `--allow-oversized-prompt` (CLI) / `allow_oversized_prompt` (MCP) as a deliberate opt-out.
+
 ## [0.61.0] - 2026-09-09
 
 ### Fixed
diff --git a/package.json b/package.json
index 5bfede5..02d855b 100644
--- a/package.json
+++ b/package.json
@@ -1,7 +1,7 @@
 {
   "name": "@mrrlin-dev/external-agents",
   "mcpName": "io.github.mrrlin-dev/external-agents",
-  "version": "0.61.0",
+  "version": "0.61.1",
   "description": "Multi-LLM MCP server for Claude Code & Codex \u2014 route AI agent work across 20+ low-cost provider buckets with round-robin dispatch and quota-aware fallback.",
   "type": "module",
   "main": "server.js",