fix(volcengine): support tiered Doubao pricing

[kunwl123456]

Summary

• add a dedicated Volcengine cost calculator that selects one request-wide tier from total prompt tokens
• restore non-zero cost tracking for Doubao Seed 2.0 Lite/Pro
• add Doubao Seed 1.8 pricing, including cache reads, input-length tiers, and its <=200 output-token discount
• display tiered model prices as min-max ranges in the model dashboard instead of $0.00

Context

The current code contains tiered price data for the Seed 2.0 models, but the cost dispatcher has no Volcengine branch, so requests fall through to flat pricing and are tracked as zero. This reintroduces the routing intent from #30357 while using Volcengine's request-wide, input-length-selected tier semantics instead of graduated bracket math.

Testing

• 22 passed — Volcengine calculator and model-cost-map unit tests on the litellm_internal_staging base
• Ruff format and lint checks passed for new Python files
• dashboard production build passed
• Prettier checks passed for changed dashboard files
• tiered dashboard transformation assertion passed

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[greptile-apps]

Greptile Summary

Adds request-wide tiered cost calculation for Volcengine models.

• Routes Volcengine token costs through a dedicated provider calculator.
• Adds Seed 1.8 pricing, cache-read accounting, and short-output rates while restoring Seed 2.0 tiered costs.
• Displays token-pricing tiers as minimum–maximum ranges in the model dashboard.
• Adds calculator, pricing-map, and dashboard transformation tests.

Confidence Score: 5/5

The PR appears safe to merge; no concrete changed-code failure was identified.

The new dispatcher preserves custom-pricing precedence, the calculator applies the repository’s established tier boundaries and flat-pricing fallback, and the dashboard transformation correctly formats the token-tier shapes introduced by this change.

Important Files Changed

Filename	Overview
litellm/cost_calculator.py	Adds the Volcengine provider dispatch after custom-pricing overrides and before generic pricing fallback.
litellm/llms/volcengine/cost_calculator.py	Implements request-wide tier selection, cached-token accounting, short-output rates, and flat-pricing fallback without an identified defect.
model_prices_and_context_window.json	Adds synchronized Seed 1.8 pricing metadata and retains the existing Seed 2.0 tier definitions.
litellm/model_prices_and_context_window_backup.json	Mirrors the canonical pricing-map changes for the bundled backup map.
tests/test_litellm/llms/volcengine/test_cost_calculator.py	Covers tier boundaries, request-wide rates, cached prompts, short outputs, and top-level dispatch.
ui/litellm-dashboard/src/app/(dashboard)/models-and-endpoints/utils/modelDataTransformer.ts	Derives formatted minimum–maximum token-price ranges from tier metadata while preserving flat-price behavior.
ui/litellm-dashboard/src/app/(dashboard)/models-and-endpoints/components/ModelsTableColumns.tsx	Labels tiered ranges appropriately while retaining the existing currency rendering.
ui/litellm-dashboard/src/components/model_dashboard/types.ts	Extends model cost fields to support formatted range strings and identifies tiered rows.

_{Reviews (1): Last reviewed commit: 4bfa7bc | Re-trigger Greptile}

[veria-ai]

Low: Higher output rate bypasses the pre-call budget bound

An authenticated user can request more than 200 output tokens while near their budget limit because _max_cost_for_cost_info() only considers output_cost_per_token and output_cost_per_reasoning_token. For this tier it reserves at 2.8e-07, while post-call accounting charges the entire output at 1.1e-06; concurrent requests can therefore incur substantially more provider spend than the atomic budget gate permits. Update the budget reservation calculation to include this conditional rate when estimating maximum output cost.

[veria-ai]

PR overview

This pull request adds support for tiered pricing for Volcengine Doubao models, including a higher conditional output-token rate.

One issue remains open in the pre-call budget calculation: it reserves output costs using the lower base rate even when the higher tier may apply. An authenticated user near their budget limit could use sufficiently large or concurrent requests to cause provider spending beyond the enforced reservation, though exploitation depends on those budget and concurrency conditions.

Open issues (1)

• Low: Higher output rate bypasses the pre-call budget bound — model_prices_and_context_window.json:45422

Fixed/addressed: 0 · PR risk: 4/10

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[codspeed-hq]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing kunwl123456:fix/volcengine-doubao-tiered-pricing (d53f574) with litellm_internal_staging (daf22ec)}