feat(models): add fireworks + together deepseek v4

[steebchen]

Supersedes #3431. Carries @vicovaro's two DeepSeek V4 provider mappings unchanged (their commits are preserved with authorship intact), rebuilt on current main, plus one capability correction found while verifying the mappings against the live providers.

What & why

Adds the two missing DeepSeek V4 provider mappings — Fireworks for V4 Pro and Together AI for V4 Flash. Both providers already serve the sibling V4 model in the catalogue, so this closes the gaps. Only packages/models/src/models/deepseek.ts changes; no gateway logic.

Model	Provider	externalId	Input / Cached / Output per 1M	Context
deepseek-v4-pro	fireworks	accounts/fireworks/models/deepseek-v4-pro	$1.74 / $0.145 / $3.48	1M
deepseek-v4-flash	together-ai	deepseek-ai/DeepSeek-V4-Flash-0731	$0.14 / $0.03 / $0.28	163K

Capability correction on top of #3431

#3431 declared the Together V4 Flash mapping tools: false, jsonOutput: false, on the basis that Together's model card lists Chat + Reasoning only. Probing the live deployment shows the card understates it:

• Tools — returns a well-formed tool_calls with finish_reason: "tool_calls"
• JSON mode (response_format: json_object) — honoured
• JSON schema (response_format: json_schema) — honoured, schema respected

This matches the together-ai V4 Pro mapping, which already declares all three. The mapping now sets tools: true, jsonOutput: true, jsonOutputSchema: true.

The practical effect of the original values was not mis-billing but silent under-routing: the gateway would have steered tool-calling and JSON traffic away from Together for this model, users pinning together-ai/deepseek-v4-flash would lose those features, and — because a false flag generates no test case — e2e would never have covered those paths.

Both mappings also had trimmed reasoningEfforts lists copied from their siblings without the siblings' explanatory comments; those are restored, per the repo convention that a non-obvious restricted list keeps its rationale.

Not changed: the Fireworks Pro mapping carries no quantization field. #3431's description called it FP8, but Fireworks' /v1/models reports no quantization and the Fireworks V4 Flash sibling doesn't declare one either, so it seemed wrong to assert it in the catalogue. Worth a follow-up if someone can confirm from the deployment.

Verification

TEST_MODELS="fireworks/deepseek-v4-pro,together-ai/deepseek-v4-flash" pnpm test:e2e

Test Files  29 passed | 2 skipped (31)
     Tests  119 passed | 91 skipped (210)

Zero failures. The correction raises the count from 110 to 119 — the 9 additional tests are exactly the Together paths the false flags had suppressed: chat-json 6→10, chat-toolcalls 4→6, chat-toolcalls-result 3→4, responses 8→10.

Also verified: both externalIds resolve upstream; per-token cost math matches the declared rates exactly (106 prompt tokens → $0.00001484 at $0.14/M, 82 completion → $0.00002296 at $0.28/M); the Fireworks priority tier resolves and applies its 1.5× multiplier in both sync and streaming; requiresDisableThinkingParam and reasoningOutput: "omit" behave correctly on Together (reasoning tokens counted, no reasoning text emitted); Fireworks streams reasoning_content as expected. packages/models catalogue specs pass (131 tests).

---

Generated by Claude Code

Summary by CodeRabbit

• New Features
  • Added Fireworks as a provider for DeepSeek V4 Pro.
  • Added Together AI as a provider for DeepSeek V4 Flash.
  • Added support for provider-specific pricing, capacity, reasoning, tool use, and JSON output capabilities.

[chatgpt-codex-connector]

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

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[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: bbaba349-c896-43c8-9fb4-49e4214321e0

📥 Commits

Reviewing files that changed from the base of the PR and between ed39fc9 and ed7c8f9.

📒 Files selected for processing (1)

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

---

Walkthrough

The model definitions add Fireworks support for DeepSeek V4 Pro and Together AI support for DeepSeek V4 Flash, including provider-specific pricing, context limits, reasoning controls, and structured output capabilities.

Changes

DeepSeek V4 provider integrations

Layer / File(s)	Summary
Provider configurations packages/models/src/models/deepseek.ts	Adds Fireworks pricing and capability settings for DeepSeek V4 Pro. Adds Together AI pricing, context, reasoning, disabled-thinking, and structured output settings for DeepSeek V4 Flash.

Estimated code review effort: 3 (Moderate) | ~15 minutes

Possibly related PRs

• theopenco/llmgateway#3431: Adds the same Fireworks and Together AI DeepSeek V4 mappings.
• theopenco/llmgateway#3429: Adds DeepSeek V4 Pro and V4 Flash provider mappings with capability and pricing settings.
• theopenco/llmgateway#3274: Adds additional DeepSeek model support for Fireworks.

Suggested reviewers: smakosh

🚥 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 main change: adding Fireworks and Together AI provider mappings for DeepSeek V4.
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 claude/pr-3431-mappings-e2e-nzjuct

---

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.}