Skip to content
Agentsmedium signalverified

A proxy that stripped one header was doubling Claude Code's API bill

Version 2.1.239 fixes streaming on Bedrock behind proxies that remove the response Content-Type header. Claude Code silently fell back to re-running every turn without streaming, and each turn was billed twice. The same release makes cost estimates show the 1.1× premium that data-residency workspaces pay.

By Redakcija WebAiRadarPublished 2 min readwritten by a model

Source

Release v2.1.239

Claude Code Releases (GitHub) · Original published August 23, 2026

Anthropic shipped Claude Code 2.1.239 on 21 August, followed by 2.1.240 and 2.1.241 on the two days after, both of which carry only bug fixes and reliability work. The one worth reading is the first, and the line worth reading twice is about money: on Amazon Bedrock, behind a proxy that strips the response Content-Type header, streaming failed in a way nobody could see. Claude Code re-ran the turn without streaming, so every turn went to the API twice and was billed twice.

The bug that cost money quietly

Nothing in the interface said anything was wrong. A turn that runs twice looks like a turn that ran once — the answer arrives, the work continues, and the difference shows up only on the invoice at the end of the month. Corporate proxies that rewrite or drop response headers are common enough that this was not an edge case for the people it hit.

The same release also fixes Claude Code hanging at startup behind an HTTPS proxy when Bedrock is used with an SSO profile and awsAuthRefresh. The credential pre-check now honours HTTPS_PROXY, which is the setting the rest of the session already respected.

What the cost figures were not saying

Cost estimates in /cost, in the status line and behind --max-budget-usd now include the 1.1× premium charged for US-only inference in data-residency workspaces. Until this release the numbers were ten percent low for exactly the customers who cannot choose where their inference runs, which is the group least able to opt out of the surcharge.

A related fix: the message shown when a monthly spend limit is exhausted now also says when the session and weekly limits reset. Knowing that the wall is monthly rather than hourly changes what you do next.

Everything else

Beyond the billing fixes, 2.1.239 carries about fifty entries. A new /claude-api upgrade migrates Python projects from the anthropic 0.x SDK to 1.x, where timeouts moved from httpx.Timeout to anthropic.Timeout. Alpine and musl builds finally load the native add-ons for image paste, clipboard and audio capture, because the binaries shipped for them are musl-built rather than glibc ones the runtime refused.

Windows gets cross-session messaging, so sessions on different machines can reach each other with SendMessage and find each other with ListAgents, as they already could on macOS and Linux. ListAgents now also tells a session its own name — the one peers use to address it — and lists live teammates rather than only subagents.

Check-ins from /goal on long-running background work now back off to 30 minutes, then an hour, then every two hours, instead of repeating every half hour.

Sources

BrandsClaude

Related

CHATGPT FOR MACiMessagereads and sends, only after you approve
Agentsmedium signal

ChatGPT now reads and sends your Apple Messages, on the Mac only

A plug-in in the ChatGPT desktop app reads and searches your iMessage, SMS and RCS threads and sends messages through Messages on your behalf. By default nothing goes out until you approve both the text and the recipients. There is a switch that removes that step, and OpenAI's own documentation argues against using it.

OpenAIverified

Screenshot of the GitHub Copilot panel inside Slack, under the heading "The new GitHub Copilot experience in Slack".
Agentsmedium signal

GitHub Copilot moved into Slack and Microsoft Teams on the same day

Both shipped on August 21 in public preview. Mention @GitHub in a channel and the agent triages issues, investigates failures, writes changes in a cloud sandbox and opens a pull request, with the conversation attached. The interesting part is not the capability list, which is familiar, but the room it moved into: the place where work gets discussed rather than written.

GitHubverified

A bobblehead figure of Nvidia's chief executive holding a game screen, above a green ARC-AGI-3 progress bar filled to 100%.
Agentsstrong signal

Nvidia's harness takes Claude Opus 5 from about 30% to a perfect ARC-AGI-3 score

Nvidia published a run in which AVO, its agent architecture, scores 100.00 RHAE on ARC-AGI-3 and clears all 183 levels across 25 environments. The same model evaluated on its own scores about 30%. In the same post Nvidia writes that the comparison is not a controlled ablation, and that sentence did not survive into the coverage.

NVIDIAverified