Add multi-layer custom_pricing for Vertex AI models by bhishmendramahala-crypto · Pull Request #715 · Portkey-AI/models

bhishmendramahala-crypto · 2026-04-15T11:35:40Z

Gemini 2.5 Pro: standard/priority/batch execution modes with <=200K/>200K context tier boundaries and grounding search prices
Gemini 2.5 Flash, Flash Lite, Flash Image: standard/priority/batch modes (flat, no context tiers)
Gemini 3.x series (3 Pro, 3 Flash, 3.1 Pro, 3.1 Flash-Lite, 3.1 Flash Image): standard/priority/batch with context tiers where applicable
Claude models (Opus 4.6, Sonnet 4.6, Haiku 4.5, plus versioned aliases): regional custom_pricing for global, us-east5, europe-west1, asia-southeast1, asia-east1
New model entries: gemma-4-26b-it, gpt-oss-20b-maas, grok-4.20-maas, grok-4.1-fast-maas, deepseek-ocr-maas, mistral-ocr-2505, veo-3.1-lite-generate-preview
Existing pricing_config + batch_config preserved as default fallback and synced with custom_pricing standard mode values
general/vertex-ai.json updated with new model entries

Description

Add multi-layer custom_pricing for Vertex AI models with execution mode support (standard/priority/batch), context tier boundaries, regional pricing, and grounding/search additional units.

New model pricing
Update existing pricing
New model configuration

Source Verification

Source Links:

Important

Please include a link to the official pricing page from the provider. Simple "I heard it from somewhere" or screenshot sources are not accepted.

Checklist

I have validated the JSON using jq or an online validator
I have verified that prices are in cents per token (not dollars)
I have included the source link above
I have signed the CLA (if first-time contributor)

Related Issue

N/A

- Gemini 2.5 Pro: standard/priority/batch execution modes with <=200K/>200K context tier boundaries and grounding search prices - Gemini 2.5 Flash, Flash Lite, Flash Image: standard/priority/batch modes (flat, no context tiers) - Gemini 3.x series (3 Pro, 3 Flash, 3.1 Pro, 3.1 Flash-Lite, 3.1 Flash Image): standard/priority/batch with context tiers where applicable - Claude models (Opus 4.6, Sonnet 4.6, Haiku 4.5, plus versioned aliases): regional custom_pricing for global, us-east5, europe-west1, asia-southeast1, asia-east1 - New model entries: gemma-4-26b-it, llama-4-scout, llama-4-maverick, gpt-oss-20b-maas, grok-4.20-maas, grok-4.1-fast-maas, deepseek-ocr-maas, mistral-ocr-2505, veo-3.1-lite-generate-preview - Existing pricing_config + batch_config preserved as default fallback - general/vertex-ai.json updated with new model entries

Copilot

Pull request overview

Adds a new custom_pricing layer to Vertex AI model pricing to support region-specific and execution-mode/context-tier pricing, while also extending the Vertex model catalog with several new models.

Changes:

Introduces custom_pricing.regions.*.execution_modes (and context tiers where applicable) for multiple Gemini and Claude models.
Normalizes many small numeric prices to scientific notation (e.g., 0.00005 → 5e-05) across pricing/vertex-ai.json.
Adds new model entries to pricing/vertex-ai.json and corresponding model type entries to general/vertex-ai.json.

Reviewed changes

Copilot reviewed 1 out of 2 changed files in this pull request and generated 9 comments.

File	Description
pricing/vertex-ai.json	Adds multi-layer `custom_pricing` (regions/execution modes/context tiers), updates numeric formatting, and appends new model pricing entries.
general/vertex-ai.json	Adds general model-type entries for newly introduced Vertex models.

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

…_write_input_token rounding - Convert all 455 scientific notation values to decimal (e.g. 5e-05 → 0.00005) - Fix batch cache_write_input_token from 0.000188 to 0.0001875 (exact 50% of standard 0.000375) Addresses PR review comments from sivadurga-d and Copilot.

…e_token in custom_pricing - Fix 22 floating point rounding artifacts (e.g. 0.00007499999999999999 → 0.000075) caused by toFixed(20) during scientific notation conversion - Add image_token to additional_units in custom_pricing for 4 image models (gemini-2.5-flash-image, gemini-3-pro-image-preview, gemini-2.5-flash-image-preview, gemini-3.1-flash-image-preview) so billing works via both response_image_token and additional_units.image_token paths

- Add asia-east1 region (1.1x multiplier) to 7 Claude models missing it: claude-sonnet-4-5-20250929, claude-sonnet-4-5@20250929, claude-opus-4-5-20251101, claude-opus-4-5@20251101, anthropic.claude-opus-4-6, claude-opus-4-6, claude-sonnet-4-6 - Add cache_write_input_token to 18 custom_pricing blocks where base config had it but custom_pricing didn't (gemini-3-pro, gemini-3-flash, gemini-3.1-pro, gemini-3.1-flash-lite, and their variants) - Add image_token to additional_units in 12 custom_pricing blocks (gemini-3-pro-preview, gemini-2.5-flash, gemini-3.1-pro-preview, gemini-2.5-flash-preview-09-2025; standard/priority get base rate, batch gets 50%) - Fix remaining scientific notation from JSON.stringify

Copilot

Pull request overview

Copilot reviewed 1 out of 2 changed files in this pull request and generated 4 comments.

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

…ding - Add cache_write_input_token to 24 context_tier blocks across gemini-3-pro-preview, gemini-3-pro-preview-02-05, gemini-3.1-pro-preview, gemini-3.1-pro-preview-customtools (all modes × both tiers) - Add image_token to 12 context_tier blocks for models with base image_token pricing (gemini-3-pro-preview, gemini-3.1-pro-preview) - Fix 45 rounded regional cache_write/cache_read values for Claude models to exact 1.1x of default (e.g. 0.000413 → 0.0004125 for Sonnet, 0.0003438 → 0.00034375 for Opus batch, 0.0000688 → 0.00006875 for Haiku)

Copilot

Pull request overview

Copilot reviewed 1 out of 2 changed files in this pull request and generated 3 comments.

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

Fix cache_write_input_token from 0.0000688 to 0.00006875 in asia-east1 batch pricing for claude-haiku-4-5-20251001 and claude-haiku-4-5@20251001, matching the exact 1.1x multiplier used in other regions.

Align all veo model pricing with Google's published per-second rates (in cents/second to match repo convention) and add resolution + audio dimensions for the pricing gaps PR code: - veo-3.1: add video_seconds_1080p=20, video_seconds_4k=40, video_audio_seconds=20 - veo-3.1-fast: fix video_seconds from 10/15 to 8 (video-only 720p), add 1080p=10, 4k=25, audio=2 - veo-3.1-lite: convert from dollars to cents (0.03→3, 0.05→5), fix audio surcharge (0.05→2), remove unused video_audio_seconds_1080p - veo-3.0: add video_seconds_1080p=20, video_audio_seconds=20 - veo-3.0-fast: fix video_seconds from 10/20 to 8, add 1080p=10, audio=2 - veo-2.0: no change (50 cents/s, 720p only, no audio) Source: https://cloud.google.com/vertex-ai/generative-ai/pricing

Fixes found during full review pass: - Fix batch cache_write_input_token to be 50% of standard (not copied from base pay_as_you_go) for 5 gemini models across 14 locations: gemini-3-pro-preview, gemini-3-pro-preview-02-05, gemini-3-flash-preview, gemini-3.1-pro-preview, gemini-3.1-pro-preview-customtools, gemini-3.1-flash-lite-preview (also fixes context_tier batch blocks) - Add missing @-notation aliases to general/vertex-ai.json: claude-sonnet-4-5@20250929, claude-haiku-4-5@20251001, claude-opus-4-5@20251101

…tex-ai.json The @ notation Claude models in general/ use the anthropic. prefix (e.g. anthropic.claude-sonnet-4-5@20250929) which were already present on main. The bare-name entries (claude-sonnet-4-5@20250929 etc.) are pricing routing aliases only and don't belong in general/.

…hanges

Copilot

Pull request overview

Copilot reviewed 1 out of 2 changed files in this pull request and generated 2 comments.

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

Copilot

Pull request overview

Copilot reviewed 1 out of 2 changed files in this pull request and generated 3 comments.

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

Ensures pricing_config (used by older gateways) has correct values matching custom_pricing standard mode, not stale/incomplete data. Fixes: - 6 wrong token prices (request, response, cache_read) - 2 wrong web_search/search prices (3.5 → 1.4) - 15 missing request_audio_token entries - 21 missing maps entries - 7 missing enterprise_web_search entries - 5 missing cache_read_input_token entries

bhishmendramahala-crypto force-pushed the pricing-update/vertex-ai-multi-layer branch from 1402fc0 to 893214c Compare April 15, 2026 11:39

bhishmendramahala-crypto requested review from Copilot and sivadurga-d April 15, 2026 12:30

Copilot started reviewing on behalf of bhishmendramahala-crypto April 15, 2026 12:30 View session

Copilot AI reviewed Apr 15, 2026

View reviewed changes

sivadurga-d reviewed Apr 16, 2026

View reviewed changes

Comment thread pricing/vertex-ai.json Outdated

Bhishmendra Mahala added 3 commits April 16, 2026 15:29

bhishmendramahala-crypto requested a review from Copilot April 16, 2026 10:23

Copilot started reviewing on behalf of bhishmendramahala-crypto April 16, 2026 10:24 View session

Copilot AI reviewed Apr 16, 2026

View reviewed changes

Comment thread pricing/vertex-ai.json

Comment thread pricing/vertex-ai.json

Comment thread pricing/vertex-ai.json

Comment thread pricing/vertex-ai.json

bhishmendramahala-crypto requested a review from Copilot April 16, 2026 10:37

Copilot started reviewing on behalf of bhishmendramahala-crypto April 16, 2026 10:37 View session

Copilot AI reviewed Apr 16, 2026

View reviewed changes

Comment thread pricing/vertex-ai.json Outdated

Comment thread pricing/vertex-ai.json Outdated

Comment thread pricing/vertex-ai.json Outdated

Bhishmendra Mahala added 5 commits April 16, 2026 16:15

fix: align haiku asia-east1 batch cache_write_input_token rounding

f48141c

Fix cache_write_input_token from 0.0000688 to 0.00006875 in asia-east1 batch pricing for claude-haiku-4-5-20251001 and claude-haiku-4-5@20251001, matching the exact 1.1x multiplier used in other regions.

restore: general/vertex-ai.json to original PR state, no formatting c…

4146045

…hanges

bhishmendramahala-crypto requested a review from Copilot April 16, 2026 11:00

Copilot started reviewing on behalf of bhishmendramahala-crypto April 16, 2026 11:01 View session

Copilot AI reviewed Apr 16, 2026

View reviewed changes

Comment thread pricing/vertex-ai.json

Comment thread pricing/vertex-ai.json

bhishmendramahala-crypto requested a review from Copilot April 16, 2026 11:19

Copilot started reviewing on behalf of bhishmendramahala-crypto April 16, 2026 11:19 View session

Copilot AI reviewed Apr 16, 2026

View reviewed changes

Comment thread pricing/vertex-ai.json

Comment thread pricing/vertex-ai.json

Comment thread pricing/vertex-ai.json

bhishmendramahala-crypto force-pushed the pricing-update/vertex-ai-multi-layer branch from 29fd8d1 to 8d31b74 Compare April 16, 2026 12:29

narengogi marked this pull request as draft April 16, 2026 19:45

Conversation

bhishmendramahala-crypto commented Apr 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Source Verification

Checklist

Related Issue

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

bhishmendramahala-crypto commented Apr 15, 2026 •

edited

Loading