fix: add 1h cache pricing for vertex claude sonnet 4.6

[biubiubiuboomboomboom]

Summary

• add cache_creation_input_token_cost_above_1hr for vertex_ai/claude-sonnet-4-6
• add the same 1h cache write rate for vertex_ai/claude-sonnet-4-6@default
• add a regression test covering Vertex Claude Sonnet 4.6 1h cache write pricing through generic_cost_per_token

Validation

• verified locally with LITELLM_LOCAL_MODEL_COST_MAP=True that generic_cost_per_token(..., custom_llm_provider="vertex_ai") returns (6.0, 0.0) for ephemeral_1h_input_tokens=1_000_000
• added a regression test for the same scenario

[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.}

[veria-ai]

Low: Pricing data addition for vertex AI model

This PR adds a cache_creation_input_token_cost_above_1hr pricing field for two existing Vertex AI Claude Sonnet 4.6 model entries in the pricing JSON, along with a corresponding test. No security-relevant changes.

---

Status: 0 open
Risk: 1/10

Posted by Veria AI · 2026-04-24T10:14:52.867Z

[biubiubiuboomboomboom]

Closing and reopening against because fork PRs to are blocked by repository policy.

[codspeed-hq]

Merging this PR will not alter performance

✅ 16 untouched benchmarks

---

_{Comparing biubiubiuboomboomboom:codex/fix-vertex-ai-claude-sonnet-4-6-cache-pricing (e14b226) with main (7b47dff)}

[greptile-apps]

Greptile Summary

This PR adds the missing cache_creation_input_token_cost_above_1hr field (6e-06) to the vertex_ai/claude-sonnet-4-6 and vertex_ai/claude-sonnet-4-6@default entries in the backup pricing JSON, and includes a regression test. The value 6e-06 is consistent with the same model's azure_ai/claude-sonnet-4-6 counterpart and with claude-sonnet-4-20250514 on the direct Anthropic provider.

Confidence Score: 5/5

Safe to merge — changes are limited to a data file addition and a clean, non-networked regression test.

Both changes are minimal and targeted: a JSON field addition with a value consistent across all equivalent provider entries, and a test that uses the local cost map with no external calls. No logic, routing, or auth code is touched.

No files require special attention.

Important Files Changed

Filename	Overview
litellm/model_prices_and_context_window_backup.json	Adds cache_creation_input_token_cost_above_1hr: 6e-06 to both vertex_ai/claude-sonnet-4-6 and vertex_ai/claude-sonnet-4-6@default; value is consistent with the model's azure_ai and anthropic pricing counterparts.
tests/test_litellm/litellm_core_utils/llm_cost_calc/test_llm_cost_calc_utils.py	Adds test_vertex_ai_claude_sonnet_4_6_1h_cache_writes_are_priced which uses the local model cost map (no real network calls) and correctly validates the new pricing field end-to-end.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["generic_cost_per_token(model='claude-sonnet-4-6', provider='vertex_ai')"] --> B["Resolve pricing key\n'vertex_ai/claude-sonnet-4-6'"]
    B --> C{"cache_creation_input_token_cost_above_1hr\npresent?"}
    C -- "Before PR: No" --> D["ephemeral_1h tokens\nbilled at 5m rate (3.75e-06)"]
    C -- "After PR: Yes (6e-06)" --> E["ephemeral_1h tokens\nbilled at 1h rate (6e-06)"]
    E --> F["prompt_cost = 1_000_000 × 6e-06 = 6.0 ✓"]

Loading

_{Reviews (1): Last reviewed commit: "fix: add vertex claude sonnet 4.6 1h cac..." | Re-trigger Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!