[Feature] Add GPT-6 Astra built-in support by Xunzhuo · Pull Request #3561 · vllm-project/semantic-router

Xunzhuo · 2026-09-08T00:51:10Z

Related #2358

Purpose

Add GPT-6 Astra as a built-in OpenAI model and make this change a complete,
copyable model-onboarding reference for future contributors.

Adds the intrinsic model card, OpenAI Chat/Responses binding, pricing,
long-context pricing tiers, unsupported request fields, always-on reasoning
family, and five official launch evaluation records.
Models the protocol difference explicitly: Chat supports low through
xhigh, while Responses additionally supports max. The new
reasoning_efforts_by_protocol field can only narrow a model binding's
validated global effort set.
Rejects unsupported Astra reasoning configurations during startup and proves
the final provider-native Chat and Responses request shapes in unit tests.
Adds a black-box E2E case to the default response-api profile. It verifies
the native model ID, Chat reasoning_effort, Responses
reasoning.effort, the Responses-only max level, removal of competing
shapes, and tool preservation at the provider boundary.
Expands the Day-0 guide with the source packet, exact files to change,
minimal user YAML, evidence rules, generated artifacts, current
impact/check/verify commands, and the named E2E command.
Regenerates the Router, CLI, built-in recipe, Dashboard, and website catalog
projections from the authored resources.

Authoritative sources:

Test Plan

make model-catalog-check
make impact ENV=cpu BASE_REF=upstream/main
make check BASE_REF=upstream/main
make verify PROFILE=response-api
make e2e-test-specific \
  E2E_PROFILE=response-api \
  E2E_TESTS=model-catalog-astra

Test Result

model-catalog-check: passed, 57 tests.
make check: passed, including pre-commit, security and architecture checks,
the complete Semantic Router suite, CLI tests (49 total; 40 passed and 9
expected skips), Dashboard lint/typecheck/frontend tests (184 files, 858
tests), Dashboard backend tests, 50 recipe conformance tests, and E2E binary
build.
model-catalog-astra: passed 1/1 in a clean Kubernetes environment
through Envoy, ExtProc, and the local provider simulator.
The complete E2E Go module passed from an exact checkout of rebased head
d78d7be1fbc8868126bc769f6bf291a43acebf89; the GitHub PR Gate is
running for this head and will be recorded here when it completes.
No external OpenAI request or credential is required by the E2E fixture.

Semantic Router PR Checklist

PR title begins with exactly one bracketed category, such as [Feature], [Bug], [Docs], [Test], [Research], [Community], or [CI/Build]
The title does not stack prefixes such as [Router][Docs]; affected modules belong in labels and the PR body
The PR links an accepted issue with exactly one owner: one wg/* label for project work or owner/maintainers for repository governance
Commits in this PR are signed off with git commit -s
The Purpose, Test Plan, and Test Result sections reflect the actual scope, commands, and blockers for this change

See CONTRIBUTING.md for the full contributor workflow and commit guidance.

netlify · 2026-09-08T00:51:15Z

✅ Deploy Preview for vllm-semantic-router ready!

Name	Link
🔨 Latest commit	`d78d7be`
🔍 Latest deploy log	https://app.netlify.com/projects/vllm-semantic-router/deploys/6a9f726be1b77d00084b6e85
😎 Deploy Preview	https://deploy-preview-3561--vllm-semantic-router.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.
🤖 Make changes	Run an agent on this branch

To edit notification comments on pull requests, go to your Netlify project configuration.

Copilot

🟡 Changes recommended

The Day-0 guide’s make impact ... CHANGED_FILES=... example references a non-existent path, so the copy/paste instructions are currently incorrect.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

This PR adds Day-0 built-in support for the OpenAI GPT-6 Astra model across the router’s catalog pipeline (authored resources → validation → generated projections), including protocol-specific reasoning-effort narrowing (Chat vs Responses) and corresponding unit + E2E coverage.

Changes:

Introduces GPT-6 Astra model card, OpenAI provider binding (pricing + restrictions), reasoning family, and 5 official launch evaluation records.
Adds reasoning_efforts_by_protocol to express per-protocol effort narrowing, plus validation and runtime materialization support.
Adds/extends contract tests and an E2E test case in the default response-api profile to assert provider-native request shapes.

File summaries

File	Description
website/static/model-catalog/catalog.json	Updates website model catalog snapshot to include GPT-6 Astra data and evaluation slots.
website/src/data/modelHubCatalogTypes.ts	Extends website binding types to include reasoning modes/efforts and per-protocol effort overrides.
website/docs/proposals/unified-model-catalog-and-evaluation-index.md	Updates proposal text/counts and documents the per-protocol effort narrowing concept.
website/docs/community/model-provider-day-0-support.md	Adds a worked Day-0 Astra example and updates generation/validation guidance.
tools/catalog/tests/test_generate_model_catalog.py	Adds catalog-generation assertions for Astra’s full authored contract and inventory inclusion.
tools/catalog/tests/test_generate_model_catalog_validation.py	Adds validation tests for protocol-scoped restrictions and `reasoning_efforts_by_protocol`.
tools/catalog/catalog_validation.py	Implements validation for `reasoning_efforts_by_protocol` and protocol-scoped API restrictions.
src/vllm-sr/cli/model_assets/latest/catalog.yaml	Regenerates CLI embedded catalog assets to include Astra.
src/semantic-router/pkg/extproc/provider_request_catalog_contract_test.go	Adds request-shape wire contract tests for Astra Chat vs Responses.
src/semantic-router/pkg/config/canonical_router_replay_test.go	Adds config materialization tests to assert Astra reasoning constraints and startup rejection behavior.
src/semantic-router/pkg/config/canonical_catalog.go	Applies `reasoning_efforts_by_protocol` when materializing provider profiles.
src/semantic-router/pkg/catalog/zz_generated_catalog.go	Regenerates router embedded catalog JSON and digest.
src/semantic-router/pkg/catalog/types.go	Adds `ReasoningEffortsByProtocol` to the Go catalog binding type.
src/semantic-router/pkg/catalog/registry.go	Ensures provider/model binding cloning deep-copies the new map field.
src/semantic-router/pkg/catalog/compiler_model_provider.go	Deep-copies `ReasoningEffortsByProtocol` when returning catalog bindings.
e2e/testcases/model_catalog_astra_day0.go	Adds black-box E2E test asserting provider-native request shapes for Astra across protocols.
e2e/profiles/response-api/values.yaml	Adds Astra aliases to the `response-api` profile and injects fixture OpenAI API key env var.
e2e/profiles/response-api/profile.go	Includes the Astra Day-0 test case in the default `response-api` profile list.
dashboard/frontend/src/types/modelCatalog.ts	Extends dashboard model catalog binding types to include `reasoning_efforts_by_protocol`.
dashboard/frontend/src/generated/modelCatalog.json	Regenerates dashboard catalog snapshot including Astra.
dashboard/backend/handlers/model_catalog_test.go	Updates dashboard catalog contract test counts for added model/evaluations.
config/recipes/built-in/latest/catalog.yaml	Regenerates built-in recipe catalog projection to include Astra resources.
config/catalog/schemas/catalog-resources-v1.schema.json	Adds schema support for `reasoning_efforts_by_protocol`.
config/catalog/resources/reasoning-families.yaml	Adds the `gpt-6-astra` reasoning family (levels + enabled-only mode).
config/catalog/resources/providers/openai.yaml	Adds Astra provider binding (protocols, pricing, restrictions, per-protocol effort narrowing).
config/catalog/resources/models/single/openai.yaml	Adds the GPT-6 Astra physical model card.
config/catalog/resources/evaluations/single/openai.yaml	Adds 5 official launch evaluation records for Astra with evidence provenance.
config/catalog/README.md	Updates baseline physical model count to reflect Astra addition.
config/catalog/catalog.yaml	Updates curated OpenAI representative models to include Astra.

Review details

Files reviewed: 26/29 changed files
Comments generated: 1
Review effort level: Lite

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

Signed-off-by: Xunzhuo Liu <xunzhuo.liu@amd.com>

nanasis · 2026-09-08T03:38:09Z

    | 'deepseek_thinking'
  reasoning_modes?: Array<'enabled' | 'disabled' | 'adaptive'>
  reasoning_efforts?: string[]
+  reasoning_efforts_by_protocol?: Record<string, string[]>


Do we need to update these files since they are related to modelCatalog?

catalogReasoningConstraints

validCatalogReasoningBindingValues

isCatalogModelBinding

Copilot AI lite review requested due to automatic review settings September 8, 2026 00:51

Xunzhuo requested review from AayushSaini101, FAUST-BENCHOU, WUKUNTAI-0211, drivebyer, ramkrishs, shraderdm, theohsiung and wilsonwu as code owners September 8, 2026 00:51

github-actions Bot added pr/needs-review Ready for reviewer attention. wg/data-plane-networking Owned by the Data Plane and Networking Workgroup. labels Sep 8, 2026

Copilot started reviewing on behalf of Xunzhuo September 8, 2026 00:51 View session

Copilot AI reviewed Sep 8, 2026

View reviewed changes

Comment thread website/docs/community/model-provider-day-0-support.md Outdated

Xunzhuo force-pushed the xunzhuo/gpt-6-astra-day0 branch 2 times, most recently from cf25087 to e5b7e04 Compare September 8, 2026 01:21

[Feature] Add GPT-6 Astra Day-0 support

4f3f27d

Signed-off-by: Xunzhuo Liu <xunzhuo.liu@amd.com>

Xunzhuo force-pushed the xunzhuo/gpt-6-astra-day0 branch from e5b7e04 to 4f3f27d Compare September 8, 2026 01:35

This was referenced Sep 8, 2026

[Feature] Add GPT-6 Astra to Azure OpenAI built-in support #3563

Open

[Feature] Add GPT-6 Astra to OpenRouter built-in support #3564

Open

[Feature] Reproduce GPT-6 Astra across the six core benchmarks #3565

Open

Xunzhuo changed the title ~~[Feature] Add GPT-6 Astra Day-0 support~~ [Feature] Add GPT-6 Astra built-in support Sep 8, 2026

test: describe Astra support accurately

d78d7be

Signed-off-by: Xunzhuo Liu <xunzhuo.liu@amd.com>

nanasis reviewed Sep 8, 2026

View reviewed changes

Navigation Menu

Uh oh!

[Feature] Add GPT-6 Astra built-in support - #3561

[Feature] Add GPT-6 Astra built-in support#3561
Xunzhuo wants to merge 2 commits into
mainfrom
xunzhuo/gpt-6-astra-day0

Xunzhuo commented Sep 8, 2026 •

edited

Loading

Uh oh!

netlify Bot commented Sep 8, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

nanasis Sep 8, 2026 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

Xunzhuo commented Sep 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Purpose

Test Plan

Test Result

Uh oh!

netlify Bot commented Sep 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

✅ Deploy Preview for vllm-semantic-router ready!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Uh oh!

Uh oh!

nanasis Sep 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Xunzhuo commented Sep 8, 2026 •

edited

Loading

netlify Bot commented Sep 8, 2026 •

edited

Loading

nanasis Sep 8, 2026 •

edited

Loading