openclaw - 💡(How to fix) Fix [Bug]: Anthropic Claude models always return "incomplete terminal response" - agent never responds on macOS

Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

Claude models always return "incomplete terminal response" on macOS - agent never responds via TUI or Telegram

Steps to reproduce

1. Install OpenClaw on macOS (npm install -g openclaw)
2. Run openclaw onboard --install-daemon
3. Configure Anthropic API key (valid, verified working via direct curl)
4. Select any Claude model (tested: claude-sonnet-4-6, claude-haiku-4-5-20251001, claude-opus-4-7)
5. Run openclaw tui
6. Type any message and press Enter
7. Agent never responds

Also reproduced via Telegram channel - same error through both interfaces.

Expected behavior

Agent responds to messages within 2-3 seconds as normal.

Actual behavior

Agent never responds. Logs show:

"anthropic/claude-sonnet-4-6 ended with an incomplete terminal response"

stopReason=stop, payloads=0, tokens=0

Error appears on every message across all models tried, both in TUI and Telegram channel. Direct curl calls to Anthropic API with the same key work perfectly.

OpenClaw version

2026.5.5 (b1abf9d) and 2026.5.7 (eeef486)

Operating system

macOS Tahoe 26.4.1 (arm64)

Install method

pm install -g openclaw

Model

• anthropic/claude-sonnet-4-6 - anthropic/claude-haiku-4-5-20251001 - anthropic/claude-opus-4-7 - anthropic/claude-3-5-sonnet-20241022 (404)

Provider / routing chain

Anthropic API key (direct) → anthropic/claude-sonnet-4-6 Also tested: anthropic/claude-haiku-4-5-20251001, anthropic/claude-opus-4-7 Auth verified working via direct curl to api.anthropic.com

Additional provider/model setup details

API key verified working with direct curl:

curl https://api.anthropic.com/v1/messages
-H "x-api-key: $ANTHROPIC_API_KEY"
-H "anthropic-version: 2023-06-01"
-H "content-type: application/json"
-d '{"model":"claude-haiku-4-5-20251001","max_tokens":10,"messages":[{"role":"user","content":"hi"}]}'

Returns valid response immediately. Issue is specific to OpenClaw routing.

Gateway logs show: thinking=medium even after /think off in TUI. /think off does not persist across /new sessions.

Logs, screenshots, and evidence

Gateway log:
[agent/embedded] empty response detected provider=anthropic/claude-sonnet-4-6 — retrying 1/1 with visible-answer continuation
[agent/embedded] empty response retries exhausted — surfacing incomplete-turn error
[agent/embedded] incomplete turn detected stopReason=stop payloads=0 — surfacing error to user
[model-fallback/decision] decision=candidate_failed reason=format code=incomplete_result detail=anthropic/claude-sonnet-4-6 ended with an incomplete terminal response

openclaw status --all output:
Gateway last log line: Embedded agent failed before reply: anthropic/claude-sonnet-4-6 ended with an incomplete terminal response
agent model: anthropic/claude-sonnet-4-6 (thinking=medium, fast=off)
Gateway: local · ws://127.0.0.1:18789 · reachable 41ms
Telegram: OK · accounts 1/1

What I tried to fix it:
- Full uninstall/reinstall (openclaw uninstall + npm install -g openclaw)
- Downgraded from 2026.5.7 to 2026.5.5
- Multiple models: claude-sonnet-4-6, claude-haiku-4-5-20251001, claude-opus-4-7
- /think off in TUI (reverts to adaptive on /new)
- openclaw doctor --fix
- pkill -9 -f openclaw + full restart
- Cleared all devices and sessions
- Mac restart

Related issues: #71880, #79896, #79893, #79888, #79886

Impact and severity

Severity: Critical - complete loss of functionality

Agent is entirely non-functional. No response is produced for any message via any channel (TUI or Telegram). OpenClaw is unusable in this state.

Fresh install on a new machine with a valid API key produces this error immediately with no workaround found after several hours of troubleshooting.

Additional information

macOS Tahoe 26.4.1 (arm64) - Apple Silicon Mac Node.js 26.0.0 OpenClaw installed via npm (npm install -g openclaw) Fresh install - no prior OpenClaw config on this machine

Multiple other users reported the same issue on May 9 2026. This appears to be a regression introduced in a recent release.

Workarounds attempted: none successful.