feat(models): add DeepSeek V4 Flash on Fireworks

[steebchen]

Summary

There was no Fireworks provider mapping for deepseek-v4-flash at all — the model was only served by deepseek, runware, novita, alibaba, deepinfra and bytedance. This adds one, pinned to the dated snapshot.

Fireworks exposes both ids:

• accounts/fireworks/models/deepseek-v4-flash — the undated alias, still resolving to the launch snapshot
• accounts/fireworks/models/deepseek-v4-flash-0731 — the current deployment

The mapping uses -0731. Verified live against the Fireworks API: the two ids return different prompt-template token counts for the same request, confirming they are distinct deployments.

Mapping details (all verified live)

• Pricing matches Fireworks' published serverless rate card: $0.14 input / $0.028 cached / $0.28 output per million.
• Priority tier is supported, but Fireworks prices DeepSeek's Priority tier at 1.5x standard — not the 1.25x that applies to the rest of its catalogue — so the mapping carries a serviceTierMultipliers override to avoid under-billing.
• Reasoning efforts: none, low, medium, high, xhigh, max. minimal is rejected with a 400 (Input should be 'low', 'medium', 'high', 'xhigh', 'max', 'none' or 'adaptive'), so it is excluded.
• Reasoning output is returned separately in reasoning_content.
• Tools, json_object and json_schema all work.
• Vision is false (Fireworks' model list reports supports_image_input: false).
• Context 1048576 per Fireworks' model list; maxOutput 393216 matching the other V4 Flash mappings.

Testing

TEST_MODELS="fireworks/deepseek-v4-flash" pnpm test:e2e — 29 files passed, 2 skipped. Confirmed with the verbose reporter that the new mapping actually ran:

• basic, /v1/chat/completions streaming
• basic reasoning
• tool calls, streaming tool calls
• JSON output, JSON output streaming
• service_tier 'priority', streaming service_tier 'priority'

Also pnpm test:unit (3577 passed), pnpm format, pnpm build.

Summary by CodeRabbit

• New Features
  • Added Fireworks support for the DeepSeek V4 Flash model.
  • Supports streaming, reasoning, tool use, JSON output, and configurable reasoning effort.
  • Supports up to 1,048,576 input tokens and 393,216 output tokens.
  • Added standard and priority service-tier pricing.
  • Enables structured output through schema-based responses.

[coderabbitai]

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: 627c53f5-2d57-41c9-bb19-caa8c3bc0ef9

📥 Commits

Reviewing files that changed from the base of the PR and between c09f941 and 3e9e087.

📒 Files selected for processing (1)

• packages/models/src/models/deepseek.ts

🚧 Files skipped from review as they are similar to previous changes (1)

• packages/models/src/models/deepseek.ts

---

Walkthrough

The model registry now includes Fireworks support for deepseek-v4-flash-0731, with pricing, priority-tier pricing, token limits, streaming, reasoning, tools, JSON output, schema output, and accepted reasoning-effort values.

Changes

DeepSeek provider configuration

Layer / File(s)	Summary
Fireworks model configuration packages/models/src/models/deepseek.ts	Adds the Fireworks deployment, pricing, priority-tier multiplier, context and output limits, capabilities, schema output support, and supported reasoning-effort values for deepseek-v4-flash.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related PRs

• theopenco/llmgateway#3072: Both changes configure reasoning-effort values for deepseek-v4-flash.
• theopenco/llmgateway#3131: Both changes modify DeepSeek provider definitions and reasoning-effort configuration.
• theopenco/llmgateway#3338: Both changes add deepseek-v4-flash-0731 for different providers.

🚥 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 and concisely describes the addition of DeepSeek V4 Flash on the Fireworks provider.
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 feat/fireworks-deepseek-v4-flash

---

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

• X
• Mastodon
• Reddit
• LinkedIn

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

[Copilot]

Pull request overview

Adds a new Fireworks provider mapping for the existing deepseek-v4-flash model definition in packages/models, using Fireworks’ dated deployment ID and capturing pricing + capability metadata so the gateway can route and bill correctly.

Changes:

• Added a fireworks provider mapping for deepseek-v4-flash, pinned to deepseek-v4-flash-0731.
• Set Fireworks pricing (including cached input) and capability flags (reasoning efforts, tools, JSON output, no vision).
• Declared Fireworks priority service tier support with a per-mapping multiplier override (1.5x) to avoid under-billing vs Fireworks’ default tier multiplier.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[coderabbitai]

Actionable comments posted: 2

🤖 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/deepseek.ts`:
- Around line 556-563: Update the Fireworks DeepSeek mapping’s parameter
declaration near reasoningEfforts to add supportedParameters for every forwarded
capability, including reasoning_effort, response_format, and tools, so the
gateway preserves these OpenAI-format fields when constructing requests.
- Line 563: Add the missing jsonOutputSchema capability flag alongside
jsonOutput in the Fireworks model mapping, then extend the provider request test
to assert that json_schema structured output is supported while preserving the
existing json_object capability.

🪄 Autofix (Beta)

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: dfa540f3-b055-421d-9f1d-40e45569146d

📥 Commits

Reviewing files that changed from the base of the PR and between 580336c and c09f941.

📒 Files selected for processing (1)

• packages/models/src/models/deepseek.ts