Skip to content

fix(models): pin DeepInfra V4 Flash to 0731 - #3384

Merged
steebchen merged 1 commit into
mainfrom
fix/deepinfra-v4-flash-0731
Aug 2, 2026
Merged

steebchen merged 1 commit into
mainfrom
fix/deepinfra-v4-flash-0731

Conversation

@steebchen

@steebchen steebchen commented Aug 2, 2026 •

Copy link
Copy Markdown
Member

Problem

The deepseek-v4-flash root model is served by seven providers. All of them except DeepInfra roll their undated model alias forward to the current 0731 snapshot automatically. DeepInfra keeps the undated deepseek-ai/DeepSeek-V4-Flash id pinned to the launch deployment and exposes the new variant only under a dated id, so requests routed to DeepInfra were silently landing on the older snapshot.

Change

Point the DeepInfra provider mapping's externalId at deepseek-ai/DeepSeek-V4-Flash-0731. No other mapping needs touching — the remaining providers already resolve to the new variant through their existing ids, and Alibaba/ByteDance/Fireworks were already pinned to their own dated ids.

Pricing, context size, quantization and capability flags are unchanged; this is purely a deployment-id correction.

Verification

  • pnpm format, full pnpm build, packages/models unit specs — all pass.
  • Full e2e suite scoped to the changed mapping: TEST_MODELS="deepinfra/deepseek-v4-flash" FULL_MODE=true pnpm test:e2e → 29 files passed, 102 tests passed.
  • Confirmed with a verbose run that the mapping actually executed live requests against DeepInfra under the new id rather than being filtered out — basic, streaming, tool calls, streaming tool calls, reasoning, and JSON output (streaming and non-streaming) all pass.

🤖 Generated with Claude Code

Summary by CodeRabbit

  • Bug Fixes
    • Updated the DeepSeek V4 Flash provider configuration to use the current external model identifier, ensuring requests are routed correctly.

DeepInfra does not roll the undated `deepseek-ai/DeepSeek-V4-Flash` alias
forward to the 0731 snapshot the way the other providers on this mapping
do, so point its externalId at the dated deployment explicitly.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Copilot AI review requested due to automatic review settings August 2, 2026 17:55
@steebchen
steebchen enabled auto-merge August 2, 2026 17:55
@chatgpt-codex-connector

Copy link
Copy Markdown

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

@coderabbitai

coderabbitai Bot commented Aug 2, 2026 •

Copy link
Copy Markdown
Contributor

Review Change Stack

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info
⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: c4b7c9a8-24a2-4890-bedc-9e30ec359ed6

📥 Commits

Reviewing files that changed from the base of the PR and between 797db4c and d44ff70.

📒 Files selected for processing (1)
  • packages/models/src/models/deepseek.ts

Walkthrough

The DeepInfra provider configuration now references deepseek-ai/DeepSeek-V4-Flash-0731 for DeepSeek V4 Flash.

Changes

DeepSeek model configuration

Layer / File(s) Summary
Update DeepInfra deployment identifier
packages/models/src/models/deepseek.ts
The DeepInfra provider now uses the dated deepseek-ai/DeepSeek-V4-Flash-0731 identifier.

Estimated code review effort: 1 (Trivial) | ~2 minutes

Possibly related PRs

Suggested reviewers: copilot, amineace

🚥 Pre-merge checks | ✅ 5
✅ Passed checks (5 passed)
Check name Status Explanation
Description Check ✅ Passed Check skipped - CodeRabbit’s high-level summary is enabled.
Title check ✅ Passed The title clearly describes pinning the DeepInfra DeepSeek V4 Flash model to the 0731 deployment.
Docstring Coverage ✅ Passed No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check ✅ Passed Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check ✅ Passed Check skipped because no linked issues were found for this pull request.
✨ Finishing Touches
📝 Generate docstrings
  • Create stacked PR
  • Commit on current branch
🧪 Generate unit tests (beta)
  • Create PR with unit tests
  • Commit unit tests in branch fix/deepinfra-v4-flash-0731

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.

❤️ Share

Comment @coderabbitai help to get the list of available commands.

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🟢 Ready to approve

The change is a minimal, targeted external deployment-id correction consistent with existing provider mapping patterns and has no apparent downstream contract or pricing impacts.

This review doesn't count toward merge requirements. Sign up for the private preview to control whether Copilot approvals count.

Pull request overview

This PR corrects the DeepInfra deployment identifier for the deepseek-v4-flash root model in packages/models, ensuring requests routed to DeepInfra hit the current 0731 snapshot rather than the older launch alias.

Changes:

  • Updated the DeepInfra provider mapping externalId for deepseek-v4-flash to use the dated DeepInfra deployment (...-0731).
File summaries
File Description
packages/models/src/models/deepseek.ts Updates DeepInfra’s externalId for deepseek-v4-flash to target the 0731 snapshot deployment.
Review details
  • Files reviewed: 1/1 changed files
  • Comments generated: 0
  • Review effort level: Lite

We're testing this review assessment. Please use 👍 or 👎 to tell us if it's correct.

@steebchen
steebchen added this pull request to the merge queue Aug 2, 2026
Merged via the queue into main with commit c11e15e Aug 2, 2026
13 checks passed
@steebchen
steebchen deleted the fix/deepinfra-v4-flash-0731 branch August 2, 2026 18:25
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants