Repository navigation
π Claude model dropdown and price table: add Fable 5.2, Fable 5.1, Haiku 5.5 and the fable alias - #18786
Conversation
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: The full list of commands accepted by this bot can be found here. DetailsNeeds approval from an approver in each of these files:Approvers can indicate their approval by writing |
Open PR overlap checkBase branch: Conflicts are checked with every PR merged onto the current tip of
|
There was a problem hiding this comment.
HUMAN DECISION NEEDED β @Danathar the PR rewrites the trusted verifier tree hash in .github/scripts/integration-governance.cjs, which is unrelated to model lists and not mentioned in the PR body. A maintainer must confirm it is intended.
.github/scripts/integration-governance.cjs:12 changes TRUSTED_SCRIPTS_TREE from 78d89054β¦ to c269cb81β¦. The in-file comment says this literal pins the executable verifier tree and must not be caller-influenced. The PR's "Files" list and changelog never mention this file. The change looks like an accidental rebase or regeneration artifact. If it is not, it re-points the trust anchor of the governance check. I could not tell whether c269cb81β¦ is a legitimate tree. It matches neither the .github/scripts tree at head (453ff867β¦) nor the one at the merge base (07f89392β¦), and the old literal matched neither base tree either. So I could not work out how it is derived. Either revert that hunk, or explain it in the PR body and show how the hash was produced.
The rest of the diff checks out. The model lists in cli_models.go, backends-tokens.js and index.html are consistent. fable is correctly kept out of the auto-heal list. The price rows and the three new tests are consistent with the stated rates. The Haiku 5.5 long-prompt undercount is documented.
β hive: agent=adjudicator-3 backend=claude model=claude-sonnet-5-5 claude=2.1.326
| // This literal is inside the enrolled .github closure. A caller-supplied | ||
| // preparedBy.sourceSHA must never choose a new executable verifier tree. | ||
| const TRUSTED_SCRIPTS_TREE = '78d89054cc569554b1066b2eb2e322a7f38c285a'; | ||
| const TRUSTED_SCRIPTS_TREE = 'c269cb813e67f341946dedd7ba9ab967a0530831'; |
There was a problem hiding this comment.
Unrelated to the PR and undisclosed: this rewrites the trusted verifier tree hash. Please revert this hunk, or explain how c269cb81β¦ was derived.
|
@Danathar this PR is back in your court ( Review asks:
Reply here or push a new head to this branch (same sign-off); CI and the review swarm re-run automatically and the PR returns to the normal path. Queued in your relay fix queue (1 waiting): fixes are handed out before new work as soon as you have capacity. |
β¦ias; price Fable 5.2 and Haiku 5.5 The pinned Claude Code 2.1.326 accepts claude-fable-5-2 (since 2.1.321), claude-fable-5-1 (since 2.1.257), claude-haiku-5-5 (since 2.1.293) and the bare `fable` alias, per the Claude Code changelog. Add them to claudePinnedCLIModels, claudeStaticModels and CLAUDE_CLI_MODELS (and the always-included aliases), record the first versions in cli-model-requirements.txt, and restore the pinned-version note the drift check reads. Add modelPrices rows for claude-fable-5-2 ($8/$40, cache read $0.20, write $10) and claude-haiku-5-5 (base rate for prompts up to 100K tokens: $0.10/$0.50, cache read $0.01, write $0.125) from Anthropic's pricing page, and bump priceTableDate. The check-cli-model-lists self-test now uses its own requirements fixture, so it no longer breaks when the real file gains a line. Refs hivecommons#18753 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WcWm7qFmFJp26r9LPdmi3t Signed-off-by: Danathar <6772335+Danathar@users.noreply.github.com>
This branch edits src/scripts (cli-model-requirements.txt and the check-cli-model-lists self-test), so the pinned tree moves to the new `git rev-parse HEAD:src/scripts`. Refs hivecommons#18753 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WcWm7qFmFJp26r9LPdmi3t Signed-off-by: Danathar <6772335+Danathar@users.noreply.github.com>
eb95a6a to
c8a80e1
Compare
There was a problem hiding this comment.
Decision needed: @clubanderson should confirm the TRUSTED_SCRIPTS_TREE re-pin in .github/scripts/integration-governance.cjs (a trust-surface literal) and the two new Anthropic price rows before this merges.
The model-list and price-row code looks correct to me; I'm routing only because the PR edits a verifier trust pin.
Verified at head c8a80e1
git rev-parse pr:src/scripts=c269cb81β¦andorigin/v5:src/scripts=78d89054β¦, so the newTRUSTED_SCRIPTS_TREEliteral (integration-governance.cjs:12) matches this branch'ssrc/scriptstree and the old value matches base. Nothing else in that file changes.claudeStaticModels,claudePinnedCLIModels,claudeAlwaysIncludeModelsandCLAUDE_CLI_MODELSreceive the same four additions in the same order; the new tests cover list membership, the allowlist cut and exact pricing.test-check-cli-model-lists.shnow uses its own fixture requirements file, so new lines in the real file no longer break the two-model cases.
Why a human
- The pin is the value a caller-supplied
sourceSHAmust never override (comment atintegration-governance.cjs:10-11). Any later edit tosrc/scriptson a competing PR will invalidate it, and the PR carriesconflicts-with-open-pr. - The price figures (Fable 5.2 $8/$40, Haiku 5.5 $0.10/$0.50) and the CLI first-version lines come from external pages I could not check. Haiku 5.5 is knowingly undercounted above 100k-token prompts (disclosed in the row comment and changelog).
β hive: agent=reviewer backend=claude model=claude-sonnet-5-5 claude=2.1.326
|
Hive status β updated in place; edits do not notify. Full history is in the Activity Monitor. Review findings from The model-list and pricing change looks correct to me; the only reason for a human is that it edits the governance verifier's pinned tree hash. Verified at
Not defects, only notes: Haiku 5.5 long-prompt spend is undercounted, which the changelog already states. Any later change to |
The claude backend's model dropdown now offers Fable 5.2, Fable 5.1, Haiku 5.5 and the bare
fablealias. The pinned Claude Code 2.1.326 accepts all four, but the hive's hand-kept lists stopped at Fable 5 and Haiku 4.5, so they were cut out even when Anthropic's live catalog listed them.The price table also gets rows for Fable 5.2 and Haiku 5.5, taken from Anthropic's pricing page. Their spend is now priced at the real rate instead of a size-tier guess.
For the operator: the four new choices show up in the agent page's claude dropdown, and cost estimates for these models stop being marked approximate.
Refs #18753
What this changes
claude-fable-5-2,claude-fable-5-1,claude-haiku-5-5and thefablealias. Nothing else in the lists changes.cli-model-requirements.txtrecords the first Claude Code version for each new id.claude-fable-5-2andclaude-haiku-5-5.priceTableDateis now 2026-11-17.Why
The allowlist exists so the dashboard never offers a model the installed CLI refuses (#8418). It was last refreshed in August, so it fell behind the CLI. Without a price row, Fable 5.2 and Haiku 5.5 spend was priced by a size-tier guess. #13488 showed how far off that guess can be.
How it was checked
Each id was checked against the Claude Code changelog up to the pinned 2.1.326:
claude-fable-5-2claude-fable-5-2)"claude-haiku-5-5claude-haiku-5-5)"claude-fable-5-1claude-fable-5-1)"fable(alias)fable; 2.1.301 fixes itTwo of the issue's hints were close but not exact. 2.1.323 only mentions
claude-haiku-5-5in a fix; the model was added in 2.1.293. 2.1.321 does addclaude-fable-5-2. Every id is accepted by 2.1.326, so nothing was left out and the pin is unchanged.Prices come from https://platform.claude.com/docs/en/about-claude/pricing (per million tokens):
The hive's own OpenRouter fixture and the Claude Code changelog give the same input, output and cache-read figures. Haiku 5.5's figures are its base rate, for prompts up to 100,000 tokens. Longer prompts bill at $0.50 / $2.50, and the table has no way to track that tier, so the row's comment says long-prompt Haiku 5.5 spend is undercounted.
Example: an agent set to
claude-fable-5-2that uses 1M tokens of each kind is now estimated at exactly $58.20. Before this change it was a tier guess.Things to know
/v1/modelscall was made to confirm the exact API order.fablealias. It is added to the always-included aliases, so every claude list offers it. It is not added as an auto-heal target, because Fable can draw from usage credits.check-cli-model-lists.shreads the pinned version from the allowlist's comment, and that note had been dropped (π± dashboard: refresh cli_models.go CLI-version commentΒ #12291), so the check always reported drift. The note is back as "2.1.326 as of π [v5] Claude model dropdown and price table: add Fable 5.2, Fable 5.1 and Haiku 5.5Β #18753". WithHIVE_MODEL_CHECK_STRICT=1the script still reports one finding, and it was there before this PR: the codex list comment says 0.166.0 while the Dockerfile pins codex 0.169.0. Fixing that needs a capture of the codex 0.169.0 catalog, which is outside this issue.TRUSTED_SCRIPTS_TREEre-pin..github/scripts/integration-governance.cjspins the git tree ofsrc/scripts(not.github/scripts): line 98 checksgit rev-parse <source>:src/scriptsagainst the literal. Any PR that editssrc/scriptsmust re-pin it, andintegration-governance.test.cjs("src/scripts tree matches the pinned TRUSTED_SCRIPTS_TREE") fails otherwise. This PR editscli-model-requirements.txtandtest-check-cli-model-lists.sh, so the literal moves from78d89054β¦(git rev-parse origin/v5:src/scripts) toc269cb81β¦(git rev-parse HEAD:src/scriptson this branch). Re-check it withgit rev-parse HEAD:src/scripts. Nothing else in that file changes.test-check-cli-model-lists.shread the real requirements file, so any new requirements line broke its two-model fixtures. It now uses its own one-line fixture.Files
src/pkg/dashboard/cli_models.go:claudeStaticModels,claudePinnedCLIModels,claudeAlwaysIncludeModelsand their comments.src/pkg/dashboard/static-src/js/backends-tokens.js:CLAUDE_CLI_MODELS.static/index.htmlwas regenerated fromstatic-srcand matches the manifest concatenation byte for byte.src/scripts/cli-model-requirements.txt: three new lines.src/scripts/test-check-cli-model-lists.sh: fixture requirements file.src/pkg/tokens/pricing.go:modelPricesrows,priceTableDate..github/scripts/integration-governance.cjs:TRUSTED_SCRIPTS_TREEre-pinned to the newsrc/scriptstree (required by the governance test).changelog.d/fixed-18753-claude-fable-haiku-5-5-models.md.Tests
TestIssue18753ClaudeListsOfferFableAndHaiku55checks that the new ids and the alias are in all three lists, with Fable newest first.TestIssue18753LiveClaudeModelsSurviveAllowlistchecks that a live/v1/modelsanswer keeps the new ids after the allowlist cut and still drops an unlisted one (claude-mythos-5-2).TestIssue18753FableAndHaiku55PricedExactlychecks that both rows resolve, that the dotted and provider-prefixed spellings map to the same row, and that a spend sample is priced exactly.gofmt -lwas run locally, as AGENTS.md requires.π€ Generated with Claude Code
https://claude.ai/code/session_01WcWm7qFmFJp26r9LPdmi3t