Repository navigation
[vscode-lm] Use measured input budgets for Opus 5.5 and Astra - #21
Merged
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What is the problem?
Opus 5.5 is missing from the current table, and Astra's advertised and accepted input budgets are stale. Using advertised limits as usable input budgets can reject requests; borrowing the earlier 197,897-token placeholder understates Opus 5.5's verified budget by about 3.4x.
How does this PR solve the problem?
Adds Opus 5.5 and updates GPT-6 Astra ("Astra 6") using session measurements from a live VS Code 1.137.0 extension host on 2026-09-23, with 62 models enumerated. Both measurements below use the copilot vendor.
Opus's configured 677,108 is a verified lower bound, not an exact ceiling. Repeated "Model declined to respond" errors prevented convergence. The earlier 935,793 advertised figure was incorrect; 871,793 is the observed advertisement. Astra's adjacent-token boundary applies to the tested prompt format, not every possible request shape. Image input and tool calling worked for both models.
Unverified/default fields: prompt caching is unverifiable through the VS Code LM API: there is no cache-hit evidence or separately exposed total-context window. The supportsPromptCache value of false is a schema-required default, not a measured finding. The contextWindow field stores the advertised input value, not a measured total-context capacity. Pricing was not measured: Astra's optional prices are omitted; Opus's zero prices are placeholders, not evidence of free usage. Neither copilotcli variant could be measured because token counting was unavailable and streams were empty.
Tests pin the measured rows and defaults while preserving main's other model updates. Only the provider table and its tests are changed. No changeset is added, consistent with recent model-table changes in this fork.
How did you test the PR?
11 passed, 0 failed, 0 skipped in one file; exit 0 (Vitest 3.2.4). Commit-hook lint and push-hook type checks each passed 10/10 tasks. Hooks warned that Node 24.15.0 differs from the requested 20.19.2. Live-probe results above come from the implementation session; they were not rerun during PR preparation.
Agent notes
Base: ac5411f (simurg79/Roo-Code main).
Scope excludes roo-vault and unrelated scratch files.