v4.0.1

Released November 3, 2026

Reduces timeouts on long‑running responses and gives Anthropic’s newest models on government deployments more room for longer, more complex answers.

Confirming this release. Very long answers, and prompts with large amounts of input, are less likely to time out. These changes are largely behind‑the‑scenes — confirm your deployment’s version with your Ask Sage administrator.

⚡ Improvements & fixes

• Fewer timeouts on long‑running responses. Ask Sage now keeps the connection active while a response is being produced, which reduces idle timeouts on very long answers and on prompts with large amounts of input.
• More room for longer answers on Anthropic 5.x models (government). On government deployments, Fable 5.1, Haiku 5.5, Opus 5.5, and Sonnet 5.5 can now produce longer responses, leaving more room for reasoning on complex prompts. Earlier Anthropic models are unchanged.
• Clearer messages when a response can’t be completed. If one of these models reaches its response limit or declines a request, you’ll see a clear explanation.

Self‑hosted and Ask Sage Edge deployments: a proxy or gateway in front of Ask Sage must pass responses through unbuffered. Otherwise the keep‑alives don’t reach clients, and long requests can still time out.

🔌 For API users

The v1 completion endpoints (/query, /query_with_file, and /query_stream) now send keep‑alives (blank lines) while a long request is in progress:

• Streaming requests receive blank lines between frames while the model works. Ignore blank lines when parsing the stream; most JSON stream parsers already do.
• Non‑streaming requests that run longer than a short wait (about a minute by default) now return HTTP 200 right away, send blank lines while the model works, and then send the usual JSON response body. Leading whitespace is valid JSON, so standard parsers still read the body. Because the status code is sent before the answer is ready, check the status field in the response body to detect an error on these longer requests rather than relying on the HTTP status code alone. This applies even on deployments that enabled the optional error‑status setting introduced in v4.0.0: once HTTP 200 has been sent, errors appear only in the body. Requests that finish within the wait are unchanged.
• Time limit. A request that runs longer than one hour (15 minutes when Live web search is on) ends with a timeout, reported in the response body as a 504 status.

🔒 Security & platform

Ask Sage services are delivered as FIPS‑hardened, digest‑pinned, and cosign‑signed images within the FedRAMP Class D certification boundary, and are continuously rebuilt against the latest patched base images as part of Ask Sage’s continuous‑monitoring commitment.