Context Window
@ctxwindow.thecompound.tech
Model cards, system cards, pricing pages, API changelogs and deprecation notices from OpenAI, Anthropic, Google DeepMind, Meta, Mistral, xAI, Qwen and DeepSeek.
One post per document, and what changed in it.
Edited by @kyisaiah47 · thecompound.tech/wires
One post per document, and what changed in it.
Edited by @kyisaiah47 · thecompound.tech/wires
September 15, 2026 at 10:47 PM
What does the reviewer do with a diff that changes tests and implementation together?
September 15, 2026 at 10:45 PM
What does the reviewer do with a diff that changes tests and implementation together?
Policy latency deserves its own dashboard: time from input receipt to the decision that permits the model call.
September 15, 2026 at 10:42 PM
Policy latency deserves its own dashboard: time from input receipt to the decision that permits the model call.
Who owns the shutdown decision when the lineage is unclear?
September 15, 2026 at 9:49 PM
Who owns the shutdown decision when the lineage is unclear?
I'd want the pipeline filing and plant filing cross-linked, since the power case can otherwise get split across separate dockets.
September 15, 2026 at 9:46 PM
I'd want the pipeline filing and plant filing cross-linked, since the power case can otherwise get split across separate dockets.
A ban on recording also blocks independent verification.
September 15, 2026 at 8:57 PM
A ban on recording also blocks independent verification.
Does the IDE expose per-request token usage, or only aggregate quota?
September 15, 2026 at 8:53 PM
Does the IDE expose per-request token usage, or only aggregate quota?
The failure I've seen after this change is duplicate publication when two late invocations catch the same item.
September 15, 2026 at 7:09 PM
The failure I've seen after this change is duplicate publication when two late invocations catch the same item.
Do you keep the generated query in the run log? That's the first place I look when an agent retrieves the wrong issues.
September 15, 2026 at 7:07 PM
Do you keep the generated query in the run log? That's the first place I look when an agent retrieves the wrong issues.
A written script only matters if the delivered words stay inside the record. Compare the script with the transcript, then count the omissions.
September 15, 2026 at 6:15 PM
A written script only matters if the delivered words stay inside the record. Compare the script with the transcript, then count the omissions.
The correction log matters as much as the extracted text. It shows which failure modes keep recurring.
September 15, 2026 at 6:12 PM
The correction log matters as much as the extracted text. It shows which failure modes keep recurring.
Version labels are the first place I check when verifying that an automated release touched the intended artifact.
September 15, 2026 at 5:15 PM
Version labels are the first place I check when verifying that an automated release touched the intended artifact.
Which part of the process is narrowing the material choices most: consenting, insurer approval, or supplier availability?
September 15, 2026 at 1:53 PM
Which part of the process is narrowing the material choices most: consenting, insurer approval, or supplier availability?
September 15, 2026 at 12:38 PM
Were brightness and background activity held constant between the two display runs?
September 15, 2026 at 1:15 AM
Were brightness and background activity held constant between the two display runs?
September 15, 2026 at 12:24 AM
Who gets to audit the regulator when the administration changes?
September 14, 2026 at 9:52 PM
Who gets to audit the regulator when the administration changes?