OpenClaw 2026.6.10: `/fast auto` mode and provider routing corrections
Introduces /fast auto, which starts short conversational turns in fast mode and drops back to normal mode for longer or fallback work while keeping the effective state visible in status. Also fixes several provider-routing bugs around Z.AI/GLM and Ollama, and adds two models to the OpenCode Go catalog.
PUBLISHED2026-06-24
OBSERVED2026-08-11
AGE1mo
SOURCES1
Automatic fast mode
- New
/fast auto: short conversational calls start fast, longer or fallback work returns to normal mode - Status reports the effective fast-mode state rather than a plain on/off or the configured value, including in connected-agent sessions
- Timing, progress and reset behaviour held consistent across retries, live model switches, fallback transitions, auto-reply and scheduled-agent runs
- Cleared Codex service-tier choice no longer carried into later runs
Providers and models
- Zhipu/GLM overload responses classified as overloads so the configured fallback is chosen for the right reason
- Bundled Z.ai GLM-5 models no longer fall through to OpenAI and emit misleading API-key errors
zai/glm-5.2thinking expanded past binary on/off; high and max map to the intended Z.AI reasoning effort/thinkmenus for live Ollama models on Telegram, Slack and Discord now show all supported levels- GLM-5.2 and Kimi K2.7 Code added to the OpenCode Go catalog with current limits;
kimi-k2.7-codecapabilities corrected to text and images (no video)
Channels
- Turn after a scheduled message retains what was delivered and whether delivery failed, without exposing cron details in the channel
- Streamed progress keeps repeated statuses that represent separate steps, and no longer sticks on stale text
COMMUNITY
No curated reactions recorded for this event. Facts and takes are kept in separate layers — community context is added by hand, never blended into the record above.