A change in Anthropic's API deployment caused a sustained spike in mid-stream ECONNRESET errors for Claude Code Max-tier users starting around July 24-25, pending Anthropic acknowledgment or fix.

state: resolvedheat: lowuncertainty: highnovelscott: lowclaude-code-api-errors anthropic-api-reliabilityAnthropic

What is this?

The web evidence is thin and largely off-topic: it documents a pattern of recurring Anthropic reliability incidents (an April 2026 outage, engineering-blog admissions about a month-long Claude Code decline tied to reasoning/config changes) and a related but distinct GitHub bug report showing Claude Code's SDK never retries connection-level errors like ECONNRESET (zero retries, falls straight to failure). None of the supplied snippets specifically confirm a July 24-25 deployment change or an Anthropic acknowledgment of this particular spike — the case's core claim is not directly corroborated by the web results, only adjacent incidents and a structural bug that would explain why such errors, whatever their cause, hit users hard with no recovery.

Why it matters to Scott

Scott builds with Claude Code and cares about coding-agent harnesses, so a reliability/retry-handling bug is in his territory, but with no wiki_hits or radar_hits to anchor it, there's no established position of his that this touches — it's just an infrastructure incident report, not something that would change what he argues or builds.

queries asked of Scott's wikis

agent harness retry logic and failure handling for API calls
coding agent reliability and error recovery patterns
vendor dependency risk in building on frontier model APIs
Claude Code as primary dev tool - dependency and outage tolerance
API reliability tracking for AI product/tooling posts

Measured heat

no measured readings yet — the hourly heat pass fills this in

How the heat travelled

no chain yet — the hourly chain pass fills this in

Evidence (4) — ⭐ canonical anchor

source	object	author	score	comments
🟠 reddit ⭐	Claude Code: sudden spike in "Connection closed mid-response" (mid-stream ECONNRESET) — anyone else since ~ Jul 24 on Max 20x? ClaudeAI	JoaoGavazzi	2	2
🟧 hn	Elevated errors on Claude Opus 5	flyaway123	48	24
🟧 hn	Elevated errors on Claude Opus 5	croemer	96	70
🟧 hn	Claude Code getting "API Error: 529 Overloaded"	croemer	3	2

Interpretation history

2026-07-27T16:25:30Z
The incident window has passed without first-party acknowledgment or independent telemetry linking the ECONNRESET spike to a deployment change. Broader overload reports remain adjacent evidence, so the specific causal hypothesis should close rather than persist on repetitive amplification.
2026-07-27T15:24:43Z
The apparent update adds no independent telemetry or first-party acknowledgment tying the mid-stream ECONNRESET spike to an Anthropic deployment; broader overload reports remain adjacent rather than causal evidence. The case stays open but is nearing a window-close if no specific corroboration appears.
2026-07-27T14:26:29Z
No new evidence connects the broader overload incidents to the specific mid-stream ECONNRESET spike or alleged deployment change; the latest activity is repetitive amplification rather than corroboration. Keep the case open for first-party acknowledgment or matching telemetry, but it remains cool.
2026-07-27T13:22:54Z
The latest activity adds no independent link between the broader Anthropic outage episode and the specific mid-stream ECONNRESET spike or alleged deployment change. With discussion merely repeating known overload evidence, the case remains open but cools pending first-party acknowledgment or matching user telemetry.
2026-07-27T12:24:08Z
Repeated status-page and overload reports establish a broader, ongoing Anthropic reliability episode, making an Anthropic-side cause more plausible. They still do not independently connect the logged mid-stream ECONNRESET spike to a deployment change or even to the reported 529 incidents, so the specific hypothesis remains uncorroborated.
2026-07-27T12:21:35Z
evidence attached: hn.story.49067964 — Repeated Claude Code overload incidents provide independent corroboration of sustained Anthropic API reliability problems.
2026-07-27T12:21:35Z
evidence attached: hn.story.49068029 — Anthropic's status report independently corroborates a broader current Claude service-reliability episode.
2026-07-27T09:22:10Z
The status-page evidence shifts this from an isolated user report toward a likely Anthropic-side reliability incident, but it does not yet confirm that the reported mid-stream ECONNRESET spike shared the same cause or resulted from a deployment change.
2026-07-27T09:20:57Z
evidence attached: hn.story.49066591 — Independent status-page evidence of elevated Claude errors materially strengthens the open case about a recent Anthropic deployment reliability problem.
2026-07-25T09:25:11Z
grounded: novel/low — Scott builds with Claude Code and cares about coding-agent harnesses, so a reliability/retry-handling bug is in his territory, but with no wiki_hits or radar_hi
2026-07-25T09:23:50Z
case created — Detailed log-based evidence of a specific, dated reliability regression in Claude Code that is independently checkable and resolvable.

Decision trace

07-28 02:25resolveThe incident window has passed without first-party acknowledgment or independent telemetry linking the ECONNRESET spike to a deployment change. Broader overload reports remain adjacent evidence, so th
07-28 02:20mark_dirtyengagement_update
07-28 01:24repriceThe apparent update adds no independent telemetry or first-party acknowledgment tying the mid-stream ECONNRESET spike to an Anthropic deployment; broader overload reports remain adjacent rather than c
07-28 01:20mark_dirtyengagement_update
07-28 00:26repriceNo new evidence connects the broader overload incidents to the specific mid-stream ECONNRESET spike or alleged deployment change; the latest activity is repetitive amplification rather than corroborat
07-28 00:20mark_dirtyengagement_update
07-28 00:20mark_dirtyengagement_update
07-27 23:22repriceThe latest activity adds no independent link between the broader Anthropic outage episode and the specific mid-stream ECONNRESET spike or alleged deployment change. With discussion merely repeating kn
07-27 23:20mark_dirtyengagement_update
07-27 23:20mark_dirtyengagement_update
07-27 23:20mark_dirtycomment_update
07-27 22:24repriceRepeated status-page and overload reports establish a broader, ongoing Anthropic reliability episode, making an Anthropic-side cause more plausible. They still do not independently connect the logged
07-27 22:21attachRepeated Claude Code overload incidents provide independent corroboration of sustained Anthropic API reliability problems.
07-27 22:21attachAnthropic's status report independently corroborates a broader current Claude service-reliability episode.
07-27 22:21propose_attachRepeated Claude Code overload incidents provide independent corroboration of sustained Anthropic API reliability problems.
07-27 22:21propose_attachAnthropic's status report independently corroborates a broader current Claude service-reliability episode.
07-27 22:20mark_dirtyengagement_update
07-27 19:22repriceThe status-page evidence shifts this from an isolated user report toward a likely Anthropic-side reliability incident, but it does not yet confirm that the reported mid-stream ECONNRESET spike shared
07-27 19:20attachIndependent status-page evidence of elevated Claude errors materially strengthens the open case about a recent Anthropic deployment reliability problem.
07-27 19:20propose_attachIndependent status-page evidence of elevated Claude errors materially strengthens the open case about a recent Anthropic deployment reliability problem.
07-25 19:25groundScott builds with Claude Code and cares about coding-agent harnesses, so a reliability/retry-handling bug is in his territory, but with no wiki_hits or radar_hits to anchor it, there's no establi
07-25 19:23createDetailed log-based evidence of a specific, dated reliability regression in Claude Code that is independently checkable and resolvable.