openclaw - 💡(How to fix) Fix [Bug]: OpenClaw Bug Report: Empty Response with stopReason=stop payloads=0 [3 comments, 2 participants]

Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

On a fresh 2026.5.7 install with a workspace migrated from 2026.4.15, every Telegram message returns "empty response retries exhausted" with stopReason=stop and payloads=0 for both anthropic/claude-sonnet-4-6 and anthropic/claude-haiku-4-5, surfacing as "All models failed (2)" and a generic error to the user.

Steps to reproduce

1. Install OpenClaw 2026.5.7 fresh (npm install -g openclaw).
2. Run openclaw setup then openclaw doctor, accepting the default to disable 46 skills with unavailable dependencies.
3. Configure model (Anthropic API key, default anthropic/claude-sonnet-4-6, picker {haiku-4-5, sonnet-4-6, opus-4-7}), web (Brave Search), channels (Telegram bot via @BotFather token).
4. Set image model: openclaw models set-image google/gemini-2.5-flash-image-preview.
5. rsync an existing workspace from a 2026.4.15 install into ~/.openclaw/workspace/ (AGENTS.md, IDENTITY.md, MEMORY.md, SOUL.md, USER.md, TOOLS.md, HEARTBEAT.md, BOOTSTRAP.md, CHANGELOG.md, plus memory/, expenses/, state/, .git/, .openclaw/).
6. Approve Telegram pairing.
7. Send any message via Telegram bot.

Expected behavior

On 2026.4.15, this same workspace produced normal Telegram replies from the agent under the same Anthropic API key family. The gateway log on that version showed agent model: anthropic/claude-... followed by user-visible replies, never incomplete turn detected: stopReason=stop payloads=0.

Actual behavior

Bot replies with either "Something went wrong while processing your request. Please try again, or use /new to start a fresh session." or "API provider returned a billing error..." (the latter despite the Anthropic console showing $99.27 remaining credit and tokens being consumed normally on this API key).

~/.openclaw/logs/gateway.err.log for every message attempt:

[agent/embedded] empty response detected: provider=anthropic/claude-sonnet-4-6 — retrying 1/1 with visible-answer continuation [agent/embedded] empty response retries exhausted: provider=anthropic/claude-sonnet-4-6 attempts=1/1 — surfacing incomplete-turn error [agent/embedded] incomplete turn detected: stopReason=stop payloads=0 — surfacing error to user [model-fallback/decision] decision=candidate_failed requested=anthropic/claude-sonnet-4-6 candidate=anthropic/claude-sonnet-4-6 reason=format code=incomplete_result [agent/embedded] empty response detected: provider=anthropic/claude-haiku-4-5 [agent/embedded] empty response retries exhausted: provider=anthropic/claude-haiku-4-5 [model-fallback/decision] fallbackStepFinalOutcome=chain_exhausted Embedded agent failed before reply: All models failed (2): anthropic/claude-sonnet-4-6 ended with an incomplete terminal response (format) | anthropic/claude-haiku-4-5 ended with an incomplete terminal response (format)

Reproduced across multiple distinct runIds (9f24f18b..., 3637b37a..., a7e6904c..., c9e70217..., 151b269e...) and two distinct sessionIds (758eb35a..., 9b9e4585...). Same outcome for primary and fallback model.

OpenClaw version

2026.5.7 (build eeef486)

Operating system

macOS 26.4.1 BuildVersion: 25E253

Install method

npm global (npm install -g openclaw)

Model

anthropic/claude-sonnet-4-6 (default), with fallback chain to anthropic/claude-haiku-4-5

Provider / routing chain

openclaw -> anthropic (direct Anthropic API; no proxy, no gateway)

Additional provider/model setup details

Auth profile anthropic:default uses an Anthropic API key generated under the Saudade workspace at console.anthropic.com (api_key=1, oauth=0, token=0). Auth store at ~/.openclaw/agents/main/agent/auth-profiles.json.

Picker: anthropic/claude-haiku-4-5, anthropic/claude-sonnet-4-6, anthropic/claude-opus-4-7. Aliases: opus -> opus-4-7, sonnet -> sonnet-4-6.

Default agent settings (per gateway log): "agent model: anthropic/claude-sonnet-4-6 (thinking=adaptive, fast=off)".

Image model: google/gemini-2.5-flash-image-preview via google:default token profile (separate path, not in the failing chain).

Logs, screenshots, and evidence

Excerpt from ~/.openclaw/logs/gateway.err.log on 2026-05-10 (multiple sessions, identical pattern):

2026-05-10T13:01:01 [agent/embedded] empty response detected: runId=9f24f18b-23d4-4d8a-9b1d-04732e3680c8 sessionId=758eb35a-d4fe-4b1a-b4dc-4ded47978b3b provider=anthropic/claude-sonnet-4-6 — retrying 1/1 with visible-answer continuation
2026-05-10T13:01:04 [agent/embedded] empty response retries exhausted: provider=anthropic/claude-sonnet-4-6 attempts=1/1 — surfacing incomplete-turn error
2026-05-10T13:01:04 [agent/embedded] incomplete turn detected: stopReason=stop payloads=0 — surfacing error to user
2026-05-10T13:01:04 [model-fallback/decision] decision=candidate_failed requested=anthropic/claude-sonnet-4-6 candidate=anthropic/claude-sonnet-4-6 reason=format code=incomplete_result errorPreview="anthropic/claude-sonnet-4-6 ended with an incomplete terminal response"
2026-05-10T13:01:07 [agent/embedded] empty response detected: provider=anthropic/claude-haiku-4-5
2026-05-10T13:01:10 [agent/embedded] empty response retries exhausted: provider=anthropic/claude-haiku-4-5 attempts=1/1
2026-05-10T13:01:10 [model-fallback/decision] fallbackStepFinalOutcome=chain_exhausted
Embedded agent failed before reply: All models failed (2): anthropic/claude-sonnet-4-6: anthropic/claude-sonnet-4-6 ended with an incomplete terminal response (format) | anthropic/claude-haiku-4-5: anthropic/claude-haiku-4-5 ended with an incomplete terminal response (format)

Reproduced across 5 distinct runIds (9f24f18b..., 3637b37a..., a7e6904c..., c9e70217..., 151b269e...) and 2 distinct sessionIds (758eb35a..., 9b9e4585...) on this install.

Anthropic console (Saudade workspace, key suffix xE6fRQAA) shows $99.27 remaining and $0.73 spent today across claude-sonnet-4-6 and claude-haiku-4-5, confirming requests reach Anthropic and tokens are billed even though no payload reaches OpenClaw.

Impact and severity

Affected: Single user of @xande_claw_bot via Telegram on this Mac mini install. Severity: Blocks all agent functionality (the agent never produces a reply). Frequency: Always. 5/5 attempted messages produced empty-response errors. Consequence: Agent is unusable; user-facing error shown to the chat is "API provider returned a billing error - your API key has run out of credits or has an insufficient balance" despite Anthropic console showing $99.27 remaining and successful token consumption on the same key. The misleading error sent to the user is itself a separate UX issue.

Additional information

Last known good version: 2026.4.15 (the old Mac running the same workspace contents replied normally via Telegram on this same bot identity). First known bad version: 2026.5.7 (current install on the new Mac mini). Workspace files were rsynced from the 2026.4.15 install into ~/.openclaw/workspace/ on 2026.5.7. No workaround currently identified. Restarting the gateway, killing stale processes, starting /new sessions, and switching between Sonnet 4.6 and Haiku 4.5 all produce the same empty-response outcome.