[Bug]: Severe slowdown, startup delay, and near-unusable responsiveness after upgrading from 2026.4.2 to 2026.4.26 on Codex setup

Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

After upgrading this Debian host from OpenClaw 2026.4.2 to 2026.4.26 in a Codex-based setup, startup became much slower and the system became laggy / nearly unusable; rolling back to 2026.4.2 restored normal behavior.

Steps to reproduce

1. Start from a working OpenClaw 2026.4.2 installation on Debian using a Codex-based setup.
2. Upgrade to OpenClaw 2026.4.26.
3. Use the system normally after upgrade and observe startup time and interactive responsiveness.
4. Observe severe slowness / lag / near-unusable behavior compared with 2026.4.2.
5. Roll back to 2026.4.2 and observe that normal behavior returns.

Expected behavior

On the same host and workflow, OpenClaw should remain reasonably responsive after upgrade, and startup time should stay close enough to the last known good version that normal use is still practical. In 2026.4.2 on this host, behavior was normal.

Actual behavior

After upgrading to 2026.4.26, startup became slow and the system became laggy / slow enough to be described as nearly unusable. After rolling back to 2026.4.2, normal behavior returned.

Observed related evidence collected during follow-up inspection:

• openclaw status on the affected environment showed the current app back on 2026.4.2, with Update available · npm update 2026.4.26.
• Logs contained repeated warnings that the config had last been written by a newer OpenClaw version: Config was last written by a newer OpenClaw (2026.4.26); current version is 2026.4.2.
• Logs during the affected period included Codex model resolution / fallback issues such as:
  • startup model warmup failed for openai-codex/gpt-5.5: Error: Unknown model: openai-codex/gpt-5.5
  • Model "openai-codex/gpt-5.5" not found. Fell back to "openrouter/openai/gpt-4o-mini".
• Logs also included repeated Codex auth refresh errors:
  • Token refresh failed: 401
  • code": "refresh_token_reused"
• Logs also showed timeouts / stuck behavior during the affected window, including:
  • gateway timeout after 30000ms
  • handshake timeout
  • stuck session

OpenClaw version

Last known good: 2026.4.2 First known bad: 2026.4.26

Operating system

Debian Linux 6.12.63+deb13-amd64 (x86_64)

Install method

npm global

Model

openai-codex/gpt-5.4 in the currently restored config; affected upgrade investigation also captured attempted resolution of openai-codex/gpt-5.5 during the bad state.

Provider / routing chain

openclaw -> openai-codex (with observed fallback to openrouter/openai/gpt-4o-mini during the bad state)

Additional provider/model setup details

This host is used with a Codex-based setup. During follow-up inspection after rollback, the restored config showed agents.defaults.model.primary = openai-codex/gpt-5.4 and plugins brave, deepseek, memory-lancedb, nextcloud-talk, openai, openrouter.

During the affected period, logs showed attempted startup warmup for openai-codex/gpt-5.5, which then failed as unknown and fell back to openrouter/openai/gpt-4o-mini. The same logs also showed repeated Codex refresh-token failures (refresh_token_reused).

I cannot prove from the available evidence whether the root cause is:

• a general 2026.4.26 regression,
• a Codex-specific regression/path interaction,
• config migration incompatibility,
• or a combination of version/config/model-auth interactions.

Logs, screenshots, and evidence

openclaw status
- Gateway: app 2026.4.2 linux 6.12.63+deb13-amd64
- Update available: npm update 2026.4.26

Representative logs:
- Config was last written by a newer OpenClaw (2026.4.26); current version is 2026.4.2.
- startup model warmup failed for openai-codex/gpt-5.5: Error: Unknown model: openai-codex/gpt-5.5
- Model "openai-codex/gpt-5.5" not found. Fell back to "openrouter/openai/gpt-4o-mini".
- Token refresh failed: 401
- code": "refresh_token_reused"
- gateway timeout after 30000ms
- handshake timeout
- stuck session

Impact and severity

Affected: this Debian host running a Codex-based OpenClaw setup Severity: High (normal use became nearly impractical after upgrade) Frequency: Observed after upgrading to 2026.4.26; rollback to 2026.4.2 restored normal behavior Consequence: Severe slowdown, slow startup, and degraded interactive usability until rollback

Additional information

Last known good version: 2026.4.2 First known bad version: 2026.4.26 Observed workaround: roll back to 2026.4.2

Release-note review did not show an explicit known regression entry for this exact symptom. Based on official release-note review plus local evidence, the most plausible areas were Codex runtime/model-path changes, config migration/version skew, diagnostics/runtime overhead, or related startup-path interactions; however, that part is inference rather than directly proven by the attached evidence.