Repository navigation
feat(models): align scx-ai-gp Qwen3.8 Max caps - #3439
Merged
smakosh merged 2 commits intoAug 6, 2026
Merged
Conversation
The scx-ai-gp mapping for qwen3.8-max was added with a narrower capability set than the Alibaba Cloud mapping for the same model, plus a quantization claim that was never verified. Drop `quantization: "fp8"` — SCX does not publish a quantization for this deployment, so asserting one is a guess that also skews the models page. Bring the model-intrinsic flags in line with the alibaba mapping, since both serve the same upstream model: `reasoningOutput: "omit"`, `reasoningMaxTokens`, `vision`, and the `supportedParameters` list that works around Qwen thinking models rejecting `tool_choice: "required"`. Deliberately not mirrored from alibaba: `webSearch`/`webSearchPrice` is Alibaba Cloud's own hosted search rather than a model capability, and `cacheReadInputPrice`/`cacheWriteInputPrice` are Alibaba's published off-ratio explicit-cache rates. Neither applies to SCX's plain OpenAI-compatible endpoint. The mapping stays `test: "skip"` — the deployment is still absent from SCX's /v1/models listing.
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
Contributor
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Repository UI Review profile: CHILL Plan: Pro Plus Run ID: 📒 Files selected for processing (1)
🚧 Files skipped from review as they are similar to previous changes (1)
WalkthroughThe SCX-AI-GP ChangesQwen3.8-Max provider configuration
Estimated code review effort: 2 (Simple) | ~10 minutes Possibly related PRs
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 1
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@packages/models/src/models/alibaba.ts`:
- Around line 1698-1703: Update the SCX provider mapping near the capabilities
fields to remove the test skip setting and add deactivatedAt with the review
date (or 2026-08-06), so scx-ai-gp is excluded from production routing.
🪄 Autofix
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Repository UI
Review profile: CHILL
Plan: Pro Plus
Run ID: 4b07098d-206b-4b7a-aafb-a301a60209b5
📒 Files selected for processing (1)
packages/models/src/models/alibaba.ts
Comment on lines
+1698
to
1703
| reasoningMaxTokens: true, | ||
| reasoningOutput: "omit", | ||
| streaming: true, | ||
| vision: true, | ||
| tools: true, | ||
| jsonOutput: true, |
Contributor
There was a problem hiding this comment.
🩺 Stability & Availability | 🟡 Minor | ⚡ Quick win
🧩 Analysis chain
🏁 Script executed:
#!/usr/bin/env bash
set -euo pipefail
# Expect: determine whether test: "skip" affects production routing or only tests.
rg -n -C 8 'test:\s*"skip"|deactivatedAt|providerId:\s*"scx-ai-gp"|qwen3\.8-max' packages --glob '*.ts'Repository: theopenco/llmgateway
Length of output: 50377
🏁 Script executed:
#!/usr/bin/env bash
set -euo pipefail
printf '--- Candidate files mentioning test/selection/deactivatedAt ---\n'
rg -n --glob '*.ts' 'type .*Test|deactivatedAt|test:\s*"skip"|skip.*provider|active provider|providers.*deactivated|isDeleted|deletedAt' packages --glob '*.ts' --max-count 200
printf '\n--- Model definition typed definitions ---\n'
fd -e ts 'types\.ts|types.*\.ts|model.*\.ts' packages --max-depth 4 | head -80 | xargs rg -n "export .*ProviderDefinition|deactivatedAt:|interface .*Provider|type .*Provider"Repository: theopenco/llmgateway
Length of output: 249
🏁 Script executed:
#!/usr/bin/env bash
set -euo pipefail
printf '%s\n' '--- Candidate files mentioning test/selection/deactivatedAt ---'
rg -n --glob '*.ts' 'type .*Test|deactivatedAt|test:\s*"skip"|skip.*provider|active provider|providers.*deactivated|isDeleted|deletedAt' packages --glob '*.ts' --max-count 200
printf '%s\n' ''
printf '%s\n' '--- Model definition typed definitions ---'
fd -e ts 'types\.ts|types.*\.ts|model.*\.ts' packages --max-depth 4 | head -80 | xargs rg -n "export .*ProviderDefinition|deactivatedAt:|interface .*Provider|type .*Provider"Repository: theopenco/llmgateway
Length of output: 30600
🏁 Script executed:
#!/usr/bin/env bash
set -euo pipefail
printf '%s\n' '--- packages/models/src/models.ts ProviderModelMapping lines 170-190 ---'
sed -n '170,190p' packages/models/src/models.ts
printf '%s\n' ''
printf '%s\n' '--- packages/actions/src/get-cheapest-model-for-provider.ts ---'
sed -n '1,80p' packages/actions/src/get-cheapest-model-for-provider.ts
printf '%s\n' ''
printf '%s\n' '--- script around active provider selection ---'
sed -n '170,195p' packages/scripts/src/export-models-dev.ts
printf '%s\n' ''
printf '%s\n' '--- model metadata provider filtering ---'
rg -n --glob '*.ts' 'ProviderModelMapping|deactivatedAt|providers\.find|providers\.filter|isMappingDeactivated|getCheapest' packages/shared packages/actions packages/scripts --max-count 200Repository: theopenco/llmgateway
Length of output: 21447
🏁 Script executed:
#!/usr/bin/env bash
set -euo pipefail
printf '%s\n' '--- select-provider-mapping.ts ---'
sed -n '1,100p' packages/actions/src/select-provider-mapping.ts
printf '%s\n' ''
printf '%s\n' '--- routes/tests mentioning test flag ---'
rg -n --glob '*.spec.ts' --glob '*.test.ts' -C 4 'test:\s*"skip"|onlyModel|skipTest|skip.*model|should not|should.*skip' packages --max-count 250Repository: theopenco/llmgateway
Length of output: 12139
🏁 Script executed:
#!/usr/bin/env bash
set -euo pipefail
printf '%s\n' '--- exact test flag consumers ---'
rg -n --glob '*.ts' --glob '!packages/**/*.spec.ts' -C 5 'providerMapping|\btest\b|\.find\(\(\s*p\s*\)\s*=>\s*.*providerId' packages --max-count 300
printf '%s\n' ''
printf '%s\n' '--- model availability/export definitions around providers fields ---'
sed -n '70,130p' packages/shared/src/components/models-directory/api-types.ts
sed -n '415,480p' packages/models/src/models.tsRepository: theopenco/llmgateway
Length of output: 50379
🏁 Script executed:
#!/usr/bin/env bash
set -euo pipefail
python3 - <<'PY'
import ast, pathlib, re
models_ts = pathlib.Path("packages/models/src/models.ts").read_text()
tree = ast.parse(models_ts)
fields = []
for node in ast.walk(tree):
if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef, ast.ClassDef)):
fields.append(node.name)
def parse_decl(text):
try:
return ast.parse(text).body[0]
except Exception:
return None
for const_name in ["ProviderMapping", "ProviderModelMapping"]:
decls = []
for name in fields:
source = getattr(tree, name, None) if isinstance(tree, ast.AST) else None
# use direct search around names
PY
printf '%s\n' '--- Locate type alias declarations by line ---'
rg -n 'const ProviderMapping|interface ProviderMapping|type ProviderMapping|interface ProviderModelMapping|type ProviderModelMapping' packages/models src --glob '*.ts'
printf '%s\n' ''
printf '%s\n' '--- Test-flag consumers excluding spec files narrowed ---'
rg -n --glob '*.ts' -C 5 '\bproviderMapping|findProviderMapping|selectProviderMapping|getCheapestModelForProvider|getCheapestFromAvailableProviders' packages/actions packages/shared packages/scripts --max-count 250Repository: theopenco/llmgateway
Length of output: 462
🏁 Script executed:
#!/usr/bin/env bash
set -euo pipefail
printf '%s\n' '--- all test flag references in non-spec TypeScript files ---'
rg -n --glob '*.ts' '!packages/**/*.spec.ts' -C 3 'test\b|skip|test:\s*|"skip"' packages/actions packages/shared packages/scripts packages/models/src --max-count 250Repository: theopenco/llmgateway
Length of output: 292
🏁 Script executed:
#!/usr/bin/env bash
set -euo pipefail
printf '%s\n' '--- all test flag references in non-spec TypeScript files ---'
rg -n --glob '*.ts' '!**/.spec.ts' -C 3 'test\b|skip|test:\s*|"skip"' packages/actions packages/shared packages/scripts packages/models/src --max-count 250 || trueRepository: theopenco/llmgateway
Length of output: 292
🏁 Script executed:
#!/usr/bin/env bash
set -euo pipefail
printf '%s\n' '--- all test flag references in non-spec TypeScript files ---'
rg -n --glob 'packages/**/*.ts' --glob '!packages/**/*.spec.ts' -C 3 'test:"skip"|test:\s*"skip"|test: "skip"|skip' packages/actions packages/shared packages/scripts packages/models --max-count 250 || true
printf '%s\n' ''
printf '%s\n' '--- provider mapping selection code ---'
sed -n '1,60p' packages/actions/src/select-provider-mapping.ts
rg -n --glob 'packages/**/*.ts' -C 3 'selectProviderMapping\(|ProviderModelMapping\[\]|\.find\(\(p\)|deactivatedAt' packages/actions packages/shared packages/scripts --max-count 200Repository: theopenco/llmgateway
Length of output: 50377
Remove test: "skip" on the SCX provider mapping.
test does not exclude a provider from production routing, only from model tests. Since scx-ai-gp returns Unsupported model, add deactivatedAt: new Date("2026-08-06") (or the review date) rather than relying on the test skip flag.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@packages/models/src/models/alibaba.ts` around lines 1698 - 1703, Update the
SCX provider mapping near the capabilities fields to remove the test skip
setting and add deactivatedAt with the review date (or 2026-08-06), so scx-ai-gp
is excluded from production routing.
Source: Coding guidelines
Adds explicit-cache pricing and web search to the scx-ai-gp mapping for qwen3.8-max, matching the capability surface of the alibaba mapping for the same upstream model. Cache: cacheReadInputPrice and cacheWriteInputPrice were previously unset, so explicit cache_control hits and cache writes both fell back to cachedInputPrice. Rates mirror the alibaba mapping. SCX's own implicit rate (0.21e-6) is kept rather than overwritten. Search: declares webSearch with the standard 0.01 per-search rate, so routing stops filtering scx-ai-gp out of web-search requests.
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.
Problem
The
scx-ai-gpmapping forqwen3.8-max(added in #3437) declares a narrower capability set than thealibabamapping for the same upstream model, and asserts a quantization that was never verified.Two consequences:
quantization: "fp8"is a guess. SCX does not publish a quantization for this deployment, and the field is surfaced on the models page.getProviderFilterReasonsfiltersscx-ai-gpout of any request using vision orreasoning_max_tokens— even though the underlying model supports both — so those requests never reach the cheaper SCX deployment.Approach
Drop
quantizationentirely rather than guessing a value.Align the model-intrinsic flags with the
alibabamapping, since both serve the same upstream model:reasoningOutput: "omit"reasoningMaxTokens: truevision: truesupportedParametersallowlist that works around Qwen thinking models rejectingtool_choice: "required"or an objectDeliberately not mirrored
Two fields on the
alibabamapping are platform features, not model capabilities, and would be wrong on SCX's plain OpenAI-compatible endpoint:webSearch/webSearchPrice— Alibaba Cloud's own hosted search (enable_search), not something the model carries with it. Declaring it would route web-search requests to a provider that cannot serve them.cacheReadInputPrice/cacheWriteInputPrice— Alibaba's published off-ratio explicit-cache rates. SCX's cache pricing is already covered bycachedInputPrice.Notes
The mapping stays
test: "skip": the deployment is still absent from SCX's/v1/modelslisting and chat completions return "Unsupported model". Capability flags are therefore declared from the upstream model's documented behavior, not from live verification — they should be re-checked once SCX brings the deployment online.Verified with
pnpm formatand a@llmgateway/modelsbuild.Summary by CodeRabbit
New Features
Bug Fixes