fix(models): update Anthropic model contextLength to 1M - #7129
Merged
diegosouzapw merged 13 commits intoJul 18, 2026
Merged
diegosouzapw merged 13 commits into
diegosouzapw merged 13 commits into
Conversation
Contributor
|
Warning You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again! |
…the default branch (diegosouzapw#7168)
… (queue_conditions alone are eligibility-only) (diegosouzapw#7179)
…uto_merge_conditions (rules-based path is EOL 2026-07-16) (diegosouzapw#7216)
…; free plan queue is serial) (diegosouzapw#7220)
…H-hosted build hang dequeued every attempt) (diegosouzapw#7225)
Owner
|
Thanks for catching that Anthropic's context windows moved — but this PR has a real routing bug as written. |
Contributor
Author
|
Thanks for the thorough review — you're absolutely right about the routing bug. Changes made1. Added
|
…copy hard-fails every PR (diegosouzapw#7341) main's copy of this test still does git I/O inside a unit test: const baseSrc = git(['show', 'origin/main:' + FILE]); Runners check out a shallow single ref, so origin/main does not resolve and the test dies with 'fatal: invalid object name origin/main'. Every PR into main fails Unit Tests (7/8) on it — today that is diegosouzapw#7313, diegosouzapw#7315, diegosouzapw#7316, diegosouzapw#7334, diegosouzapw#7336 and diegosouzapw#7337, six PRs red on a defect none of them introduced. diegosouzapw#7313 has no other red at all. release/v3.8.49 already carries a fix (2e42b8e, diegosouzapw#7174: try/catch, fetch origin/main on demand, t.skip() when unreachable), but it only reaches main at release time — so main stays broken for the whole cycle. Cherry-picking it would also import a new problem: PR Test Policy classifies t.skip() as a silenced assertion, which we watched it correctly catch on diegosouzapw#7300 today. This is the hermetic version instead (ported from diegosouzapw#7327, which does the same for the release branch): read the file straight off disk, compare against an empty base so baseTaut/baseExtTaut are 0 — the strictest possible comparison point — and call evaluateMasking() directly. No git ref, no fetch, no skip, nothing the runner's checkout depth can break. The diegosouzapw#6634 regression stays covered: the guard's logic lives in SELF_TEST_FIXTURE_RE (check-test-masking.mjs:337), not in the test. Proven both ways on main before committing — neutralise SELF_TEST_FIXTURE_RE to /$^/ and the test FAILS; restore it and it passes 2/2, with check-test-masking.mjs left byte-identical. Co-authored-by: growab <nekron@icloud.com>
…bers (diegosouzapw#7347) main's ratchet had been failing --require-tighten on every PR: 11 metrics improved but the baseline was never tightened. Same class as the diegosouzapw#6634 selfref guard — an infra fix that lands only on the release branch leaves main red for the whole cycle, and every PR into main pays for it. Values are the merged-coverage numbers from a run on main itself (a local run measures ~68% vs CI's ~80%; the baseline's own note warns about that gap). Only the 11 coverage values change — gitleaks and semgrepFindings keep main's own state. No changelog fragment: diegosouzapw#7326 carries it on release/v3.8.49, and a second one here would double the entry at release time.
Opus 4.6, Sonnet 4.6, and Sonnet 5 all have 1M context window since GA (2026-03-13). Update: - agyModels + antigravityModelAliases: 200000 -> 1048576 (binary 1M, matches existing convention in those files) - claude registry: 200000 -> 1000000 (decimal 1M, matches other entries in that file) Older models (opus-4-5, sonnet-4-5, haiku-4-5) and defaultContextLength: 200000 left untouched. Signed-off-by: Minxi Hou <houminxi@gmail.com>
Sonnet 4.6 has 1M context GA since 2026-02-17. Without this entry, the CC-compatible wire image omits the context-1m beta header for Sonnet 4.6 requests, causing large-context requests to be rejected by Anthropic.
Verify that every Claude model with contextLength > 200K has a matching entry in the beta header allowlist, and vice versa. Parses source files directly (no module imports needed). Signed-off-by: Minxi Hou <houminxi@gmail.com>
Add sanity check that parser finds at least one model. Handle type annotations in CONTEXT_1M_SUPPORTED_MODELS assignment. Remove weaker subagent test. Signed-off-by: Minxi Hou <houminxi@gmail.com>
HouMinXi
force-pushed
the
fix/claude-context-length
branch
from
July 16, 2026 11:27
be8627d to
9e0f0d2
Compare
# Conflicts: # .mergify.yml
…4.6 fix Three pre-existing tests hardcoded the old 200000 contextLength for claude-opus-4-6-thinking / claude-sonnet-4-6 / claude-sonnet-5, which this PR's own registry change (agyModels.ts, antigravityModelAliases.ts, claude registry) legitimately bumped to 1M GA: - antigravity-model-aliases.test.ts: deepEqual assertions for claude-opus-4-6-thinking and claude-sonnet-5 in the Antigravity catalog expected contextLength: 200000; now 1048576. Added an explicit contextLength assertion for claude-sonnet-4-6 too. - auto-combo-context-advertising.test.ts: resolveComboContextLimit regression test asserted the claude target's own limit as 200000; the test's intent (own limit wins over an 8k sibling, not compressed) is unchanged, only the registry's current correct value (1000000) needed updating. Also refreshed a stale comment in the MAX-of-candidates test (functionally unaffected, since gemini's 1048576 already wins either way). - models-catalog-route.test.ts: refreshed a stale inline comment (the actual assertion checks the MIN across combo targets, 128000, unaffected by the 1M bump). Verified via Anthropic's own docs (platform.claude.com/docs/en/build-with-claude/context-windows): "Claude Opus 4.8, Claude Opus 4.7, Claude Opus 4.6, Claude Sonnet 5, and Claude Sonnet 4.6 have a 1M-token context window ... on the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry" -- matches this PR's claimed 2026-03-13 GA date exactly, and Google Cloud coverage corroborates the Antigravity-hosted ids. Independently corroborated for the Antigravity-specific path by unrelated third-party projects fixing the same gap (earendil-works/pi#2209, earendil-works/pi#2194). Swept the full test suite for other contextLength/contextWindow assertions against these three model ids (grep + targeted runs across models-catalog-route, model-capabilities-registry, t31-t33-t34-t38-model-specs, executor-antigravity, agy-provider, provider-models-config, model-metadata-registry, claude-web-sonnet5-registry, combo-routing-engine, command-code-executor, combo-lockout-quota-reset -- 157 tests green); the Bedrock-hosted registry and modelSpecs.ts module were already correctly at 1000000 for these models, reinforcing this PR's direction. Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
Owner
|
Thanks @HouMinXi! Merged into release/v3.8.49 after merge-train validation on the combined tree. The fallback-chain try/catch was split into the proFallbackChain module to clear the complexity gates — authorship preserved. |
diegosouzapw
added a commit
that referenced
this pull request
Jul 19, 2026
#7129 added claude-sonnet-4-6 to CONTEXT_1M_SUPPORTED_MODELS (1M context GA'd 2026-02-17) but missed this test in its sweep. A non-CC anthropic-compatible target with extendedContext:true now legitimately receives the context-1m beta header for this model — updating the stale undefined expectation.
This was referenced Jul 22, 2026
Merged
HouMinXi
added a commit
to HouMinXi/OmniRoute
that referenced
this pull request
Aug 2, 2026
…#7129) * chore(ci): add .mergify.yml to main — Mergify only reads config from the default branch (diegosouzapw#7168) * fix(ci): add the auto-enqueue pull_request_rule to the Mergify config (queue_conditions alone are eligibility-only) (diegosouzapw#7179) * fix(ci): migrate Mergify auto-enqueue to merge_protections_settings.auto_merge_conditions (rules-based path is EOL 2026-07-16) (diegosouzapw#7216) * fix(ci): drop Mergify batch settings (batching is a paid-tier feature; free plan queue is serial) (diegosouzapw#7220) * fix(ci): merge queue tolerates the advisory dast-smoke failure (its GH-hosted build hang dequeued every attempt) (diegosouzapw#7225) * test(ci): make the diegosouzapw#6634 selfref guard hermetic — main's copy hard-fails every PR (diegosouzapw#7341) main's copy of this test still does git I/O inside a unit test: const baseSrc = git(['show', 'origin/main:' + FILE]); Runners check out a shallow single ref, so origin/main does not resolve and the test dies with 'fatal: invalid object name origin/main'. Every PR into main fails Unit Tests (7/8) on it — today that is diegosouzapw#7313, diegosouzapw#7315, diegosouzapw#7316, diegosouzapw#7334, diegosouzapw#7336 and diegosouzapw#7337, six PRs red on a defect none of them introduced. diegosouzapw#7313 has no other red at all. release/v3.8.49 already carries a fix (8bbd411, diegosouzapw#7174: try/catch, fetch origin/main on demand, t.skip() when unreachable), but it only reaches main at release time — so main stays broken for the whole cycle. Cherry-picking it would also import a new problem: PR Test Policy classifies t.skip() as a silenced assertion, which we watched it correctly catch on diegosouzapw#7300 today. This is the hermetic version instead (ported from diegosouzapw#7327, which does the same for the release branch): read the file straight off disk, compare against an empty base so baseTaut/baseExtTaut are 0 — the strictest possible comparison point — and call evaluateMasking() directly. No git ref, no fetch, no skip, nothing the runner's checkout depth can break. The diegosouzapw#6634 regression stays covered: the guard's logic lives in SELF_TEST_FIXTURE_RE (check-test-masking.mjs:337), not in the test. Proven both ways on main before committing — neutralise SELF_TEST_FIXTURE_RE to /$^/ and the test FAILS; restore it and it passes 2/2, with check-test-masking.mjs left byte-identical. Co-authored-by: growab <nekron@icloud.com> * chore(quality): tighten main's coverage baseline to the CI's real numbers (diegosouzapw#7347) main's ratchet had been failing --require-tighten on every PR: 11 metrics improved but the baseline was never tightened. Same class as the diegosouzapw#6634 selfref guard — an infra fix that lands only on the release branch leaves main red for the whole cycle, and every PR into main pays for it. Values are the merged-coverage numbers from a run on main itself (a local run measures ~68% vs CI's ~80%; the baseline's own note warns about that gap). Only the 11 coverage values change — gitleaks and semgrepFindings keep main's own state. No changelog fragment: diegosouzapw#7326 carries it on release/v3.8.49, and a second one here would double the entry at release time. * fix(models): update Anthropic model contextLength to 1M Opus 4.6, Sonnet 4.6, and Sonnet 5 all have 1M context window since GA (2026-03-13). Update: - agyModels + antigravityModelAliases: 200000 -> 1048576 (binary 1M, matches existing convention in those files) - claude registry: 200000 -> 1000000 (decimal 1M, matches other entries in that file) Older models (opus-4-5, sonnet-4-5, haiku-4-5) and defaultContextLength: 200000 left untouched. Signed-off-by: Minxi Hou <houminxi@gmail.com> * fix(claude): add sonnet-4-6 to CONTEXT_1M_SUPPORTED_MODELS Sonnet 4.6 has 1M context GA since 2026-02-17. Without this entry, the CC-compatible wire image omits the context-1m beta header for Sonnet 4.6 requests, causing large-context requests to be rejected by Anthropic. * test: regression test for CONTEXT_1M_SUPPORTED_MODELS allowlist Verify that every Claude model with contextLength > 200K has a matching entry in the beta header allowlist, and vice versa. Parses source files directly (no module imports needed). Signed-off-by: Minxi Hou <houminxi@gmail.com> * test: improve regression test with empty-registry guard Add sanity check that parser finds at least one model. Handle type annotations in CONTEXT_1M_SUPPORTED_MODELS assignment. Remove weaker subagent test. Signed-off-by: Minxi Hou <houminxi@gmail.com> * test: update stale contextLength expectations for the 1M Sonnet/Opus 4.6 fix Three pre-existing tests hardcoded the old 200000 contextLength for claude-opus-4-6-thinking / claude-sonnet-4-6 / claude-sonnet-5, which this PR's own registry change (agyModels.ts, antigravityModelAliases.ts, claude registry) legitimately bumped to 1M GA: - antigravity-model-aliases.test.ts: deepEqual assertions for claude-opus-4-6-thinking and claude-sonnet-5 in the Antigravity catalog expected contextLength: 200000; now 1048576. Added an explicit contextLength assertion for claude-sonnet-4-6 too. - auto-combo-context-advertising.test.ts: resolveComboContextLimit regression test asserted the claude target's own limit as 200000; the test's intent (own limit wins over an 8k sibling, not compressed) is unchanged, only the registry's current correct value (1000000) needed updating. Also refreshed a stale comment in the MAX-of-candidates test (functionally unaffected, since gemini's 1048576 already wins either way). - models-catalog-route.test.ts: refreshed a stale inline comment (the actual assertion checks the MIN across combo targets, 128000, unaffected by the 1M bump). Verified via Anthropic's own docs (platform.claude.com/docs/en/build-with-claude/context-windows): "Claude Opus 4.8, Claude Opus 4.7, Claude Opus 4.6, Claude Sonnet 5, and Claude Sonnet 4.6 have a 1M-token context window ... on the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry" -- matches this PR's claimed 2026-03-13 GA date exactly, and Google Cloud coverage corroborates the Antigravity-hosted ids. Independently corroborated for the Antigravity-specific path by unrelated third-party projects fixing the same gap (earendil-works/pi#2209, earendil-works/pi#2194). Swept the full test suite for other contextLength/contextWindow assertions against these three model ids (grep + targeted runs across models-catalog-route, model-capabilities-registry, t31-t33-t34-t38-model-specs, executor-antigravity, agy-provider, provider-models-config, model-metadata-registry, claude-web-sonnet5-registry, combo-routing-engine, command-code-executor, combo-lockout-quota-reset -- 157 tests green); the Bedrock-hosted registry and modelSpecs.ts module were already correctly at 1000000 for these models, reinforcing this PR's direction. Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> --------- Signed-off-by: Minxi Hou <houminxi@gmail.com> Co-authored-by: Diego Rodrigues de Sa e Souza <8016841+diegosouzapw@users.noreply.github.com> Co-authored-by: growab <nekron@icloud.com> Co-authored-by: Diego Rodrigues de Sa e Souza <diegosouza.pw@gmail.com>
HouMinXi
pushed a commit
to HouMinXi/OmniRoute
that referenced
this pull request
Aug 2, 2026
…zapw#7129) diegosouzapw#7129 added claude-sonnet-4-6 to CONTEXT_1M_SUPPORTED_MODELS (1M context GA'd 2026-02-17) but missed this test in its sweep. A non-CC anthropic-compatible target with extendedContext:true now legitimately receives the context-1m beta header for this model — updating the stale undefined expectation.
muhamadgalihsaputra
pushed a commit
to niyatna/NiyatnaRoute
that referenced
this pull request
Sep 27, 2026
…#7129) * chore(ci): add .mergify.yml to main — Mergify only reads config from the default branch (diegosouzapw#7168) * fix(ci): add the auto-enqueue pull_request_rule to the Mergify config (queue_conditions alone are eligibility-only) (diegosouzapw#7179) * fix(ci): migrate Mergify auto-enqueue to merge_protections_settings.auto_merge_conditions (rules-based path is EOL 2026-07-16) (diegosouzapw#7216) * fix(ci): drop Mergify batch settings (batching is a paid-tier feature; free plan queue is serial) (diegosouzapw#7220) * fix(ci): merge queue tolerates the advisory dast-smoke failure (its GH-hosted build hang dequeued every attempt) (diegosouzapw#7225) * test(ci): make the diegosouzapw#6634 selfref guard hermetic — main's copy hard-fails every PR (diegosouzapw#7341) main's copy of this test still does git I/O inside a unit test: const baseSrc = git(['show', 'origin/main:' + FILE]); Runners check out a shallow single ref, so origin/main does not resolve and the test dies with 'fatal: invalid object name origin/main'. Every PR into main fails Unit Tests (7/8) on it — today that is diegosouzapw#7313, diegosouzapw#7315, diegosouzapw#7316, diegosouzapw#7334, diegosouzapw#7336 and diegosouzapw#7337, six PRs red on a defect none of them introduced. diegosouzapw#7313 has no other red at all. release/v3.8.49 already carries a fix (83a7551, diegosouzapw#7174: try/catch, fetch origin/main on demand, t.skip() when unreachable), but it only reaches main at release time — so main stays broken for the whole cycle. Cherry-picking it would also import a new problem: PR Test Policy classifies t.skip() as a silenced assertion, which we watched it correctly catch on diegosouzapw#7300 today. This is the hermetic version instead (ported from diegosouzapw#7327, which does the same for the release branch): read the file straight off disk, compare against an empty base so baseTaut/baseExtTaut are 0 — the strictest possible comparison point — and call evaluateMasking() directly. No git ref, no fetch, no skip, nothing the runner's checkout depth can break. The diegosouzapw#6634 regression stays covered: the guard's logic lives in SELF_TEST_FIXTURE_RE (check-test-masking.mjs:337), not in the test. Proven both ways on main before committing — neutralise SELF_TEST_FIXTURE_RE to /$^/ and the test FAILS; restore it and it passes 2/2, with check-test-masking.mjs left byte-identical. Co-authored-by: growab <nekron@icloud.com> * chore(quality): tighten main's coverage baseline to the CI's real numbers (diegosouzapw#7347) main's ratchet had been failing --require-tighten on every PR: 11 metrics improved but the baseline was never tightened. Same class as the diegosouzapw#6634 selfref guard — an infra fix that lands only on the release branch leaves main red for the whole cycle, and every PR into main pays for it. Values are the merged-coverage numbers from a run on main itself (a local run measures ~68% vs CI's ~80%; the baseline's own note warns about that gap). Only the 11 coverage values change — gitleaks and semgrepFindings keep main's own state. No changelog fragment: diegosouzapw#7326 carries it on release/v3.8.49, and a second one here would double the entry at release time. * fix(models): update Anthropic model contextLength to 1M Opus 4.6, Sonnet 4.6, and Sonnet 5 all have 1M context window since GA (2026-03-13). Update: - agyModels + antigravityModelAliases: 200000 -> 1048576 (binary 1M, matches existing convention in those files) - claude registry: 200000 -> 1000000 (decimal 1M, matches other entries in that file) Older models (opus-4-5, sonnet-4-5, haiku-4-5) and defaultContextLength: 200000 left untouched. Signed-off-by: Minxi Hou <houminxi@gmail.com> * fix(claude): add sonnet-4-6 to CONTEXT_1M_SUPPORTED_MODELS Sonnet 4.6 has 1M context GA since 2026-02-17. Without this entry, the CC-compatible wire image omits the context-1m beta header for Sonnet 4.6 requests, causing large-context requests to be rejected by Anthropic. * test: regression test for CONTEXT_1M_SUPPORTED_MODELS allowlist Verify that every Claude model with contextLength > 200K has a matching entry in the beta header allowlist, and vice versa. Parses source files directly (no module imports needed). Signed-off-by: Minxi Hou <houminxi@gmail.com> * test: improve regression test with empty-registry guard Add sanity check that parser finds at least one model. Handle type annotations in CONTEXT_1M_SUPPORTED_MODELS assignment. Remove weaker subagent test. Signed-off-by: Minxi Hou <houminxi@gmail.com> * test: update stale contextLength expectations for the 1M Sonnet/Opus 4.6 fix Three pre-existing tests hardcoded the old 200000 contextLength for claude-opus-4-6-thinking / claude-sonnet-4-6 / claude-sonnet-5, which this PR's own registry change (agyModels.ts, antigravityModelAliases.ts, claude registry) legitimately bumped to 1M GA: - antigravity-model-aliases.test.ts: deepEqual assertions for claude-opus-4-6-thinking and claude-sonnet-5 in the Antigravity catalog expected contextLength: 200000; now 1048576. Added an explicit contextLength assertion for claude-sonnet-4-6 too. - auto-combo-context-advertising.test.ts: resolveComboContextLimit regression test asserted the claude target's own limit as 200000; the test's intent (own limit wins over an 8k sibling, not compressed) is unchanged, only the registry's current correct value (1000000) needed updating. Also refreshed a stale comment in the MAX-of-candidates test (functionally unaffected, since gemini's 1048576 already wins either way). - models-catalog-route.test.ts: refreshed a stale inline comment (the actual assertion checks the MIN across combo targets, 128000, unaffected by the 1M bump). Verified via Anthropic's own docs (platform.claude.com/docs/en/build-with-claude/context-windows): "Claude Opus 4.8, Claude Opus 4.7, Claude Opus 4.6, Claude Sonnet 5, and Claude Sonnet 4.6 have a 1M-token context window ... on the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry" -- matches this PR's claimed 2026-03-13 GA date exactly, and Google Cloud coverage corroborates the Antigravity-hosted ids. Independently corroborated for the Antigravity-specific path by unrelated third-party projects fixing the same gap (earendil-works/pi#2209, earendil-works/pi#2194). Swept the full test suite for other contextLength/contextWindow assertions against these three model ids (grep + targeted runs across models-catalog-route, model-capabilities-registry, t31-t33-t34-t38-model-specs, executor-antigravity, agy-provider, provider-models-config, model-metadata-registry, claude-web-sonnet5-registry, combo-routing-engine, command-code-executor, combo-lockout-quota-reset -- 157 tests green); the Bedrock-hosted registry and modelSpecs.ts module were already correctly at 1000000 for these models, reinforcing this PR's direction. Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com> --------- Signed-off-by: Minxi Hou <houminxi@gmail.com> Co-authored-by: Diego Rodrigues de Sa e Souza <8016841+diegosouzapw@users.noreply.github.com> Co-authored-by: growab <nekron@icloud.com> Co-authored-by: Diego Rodrigues de Sa e Souza <diegosouza.pw@gmail.com>
muhamadgalihsaputra
pushed a commit
to niyatna/NiyatnaRoute
that referenced
this pull request
Sep 27, 2026
…zapw#7129) diegosouzapw#7129 added claude-sonnet-4-6 to CONTEXT_1M_SUPPORTED_MODELS (1M context GA'd 2026-02-17) but missed this test in its sweep. A non-CC anthropic-compatible target with extendedContext:true now legitimately receives the context-1m beta header for this model — updating the stale undefined expectation.
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
2 participants
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.
Opus 4.6, Sonnet 4.6, and Sonnet 5 all have 1M context window since GA (2026-03-13).
Changes:
Older models (4.5 series) and defaultContextLength left untouched.