fix: add Bedrock models publishers/anthropic/models/claude-opus-4-8 +3 more

[github-actions]

fix: add Bedrock models publishers/anthropic/models/claude-opus-4-8 +3 more

Closes #671

Source issue: #671

Summary

Field	Value
Provider	bedrock
Primary model	publishers/anthropic/models/claude-opus-4-8
Changed models	publishers/anthropic/models/claude-opus-4-8 anthropic.claude-opus-4-8 global.anthropic.claude-opus-4-8 us.anthropic.claude-opus-4-8
Added models	publishers/anthropic/models/claude-opus-4-8 anthropic.claude-opus-4-8 global.anthropic.claude-opus-4-8 us.anthropic.claude-opus-4-8
Updated models	None
Verification sources	1 2 3

Verified metadata

Model	Display name	Parent	Providers	Format	Flavor	Token limits	Pricing	Lifecycle
publishers/anthropic/models/claude-opus-4-8	Claude 4.8 Opus		vertex	anthropic	chat	input=1000000, output=128000	in/out=5/25 per 1M; cache read=0.5 per 1M; cache write=6.25 per 1M	multimodal=true; reasoning=true
anthropic.claude-opus-4-8	Claude 4.8 Opus		bedrock	anthropic	chat	input=1000000, output=128000	in/out=5/25 per 1M; cache read=0.5 per 1M; cache write=6.25 per 1M	multimodal=true; reasoning=true
global.anthropic.claude-opus-4-8	Global Claude 4.8 Opus	anthropic.claude-opus-4-8	bedrock	anthropic	chat	input=1000000, output=128000	in/out=5/25 per 1M; cache read=0.5 per 1M; cache write=6.25 per 1M	parent=anthropic.claude-opus-4-8; multimodal=true; reasoning=true
us.anthropic.claude-opus-4-8	US Claude 4.8 Opus	anthropic.claude-opus-4-8	bedrock	anthropic	chat	input=1000000, output=128000	in/out=5.5/27.5 per 1M; cache read=0.55 per 1M; cache write=6.875 per 1M	parent=anthropic.claude-opus-4-8; multimodal=true; reasoning=true

Verification notes

Verification

Official source verification

• https://platform.claude.com/docs/en/docs/about-claude/models/all-models: Confirmed model IDs (anthropic.claude-opus-4-8 for Bedrock, claude-opus-4-8 for Vertex AI), context window (1M tokens), max output (128k tokens), multimodal (vision) support, adaptive thinking support, pricing ($5 input / $25 output per MTok).
• https://platform.claude.com/docs/en/about-claude/pricing: Confirmed pricing: base input $5/MTok, output $25/MTok, 5m cache write $6.25/MTok (1.25x), 1h cache write $10/MTok (2x), cache read $0.50/MTok (0.1x). Confirmed 10% regional premium for Bedrock/Vertex regional endpoints on Claude 4.5+ models.
• https://platform.claude.com/docs/en/build-with-claude/claude-on-vertex-ai: Confirmed Vertex AI model ID claude-opus-4-8, global/multi-region/regional endpoint types. Regional and multi-region carry 10% premium over global. Docs state "model availability varies by region" but do not list specific supported regions for Claude models.

sync_models cross-check

sync_models reference: https://raw.githubusercontent.com/BerriAI/litellm/refs/heads/main/litellm/model_prices_and_context_window_backup.json

anthropic.claude-opus-4-8 (sync_models provider: bedrock_converse):

• max_input_tokens: 1000000 — matches proposed ✓
• max_output_tokens: 128000 — matches proposed ✓
• input cost: $5/MTok (5e-6 per token) — matches proposed ✓
• output cost: $25/MTok (2.5e-5 per token) — matches proposed ✓
• cache write: $6.25/MTok (6.25e-6 per token) — matches proposed ✓
• cache read: $0.50/MTok (5e-7 per token) — matches proposed ✓

global.anthropic.claude-opus-4-8 (sync_models provider: bedrock_converse):

• All fields identical to base entry — matches proposed ✓

us.anthropic.claude-opus-4-8 (sync_models provider: bedrock_converse):

• input cost: $5.50/MTok (5.5e-6 per token) — matches proposed ✓
• output cost: $27.50/MTok (2.75e-5 per token) — matches proposed ✓
• cache write: $6.875/MTok (6.875e-6 per token) — matches proposed ✓
• cache read: $0.55/MTok (5.5e-7 per token) — matches proposed ✓

publishers/anthropic/models/claude-opus-4-8 (Vertex):

• No entry found in sync_models for vertex variant. Pricing verified from Anthropic pricing page and matches existing publishers/anthropic/models/claude-opus-4-7 catalog entry.

Deviations from sync_models

No deviations. All proposed numeric values match sync_models exactly for the three Bedrock variants. The Vertex variant has no sync_models entry to compare against.

Fields not published or not applicable

• supported_regions for the Vertex model: Omitted. The Anthropic Vertex AI documentation states "model availability varies by region" but does not publish a specific region list for Claude models. The Google locations page referenced in the instructions covers Google-owned models (Gemini), not partner models (Anthropic). No official region list found; field omitted per instructions.
• deprecation / deprecation_date: Not applicable; Claude Opus 4.8 is the current latest Opus model with no announced deprecation.
• input_cache_write_5m_cost_per_mil_tokens and input_cache_write_1h_cost_per_mil_tokens for Bedrock variants: Omitted for Bedrock entries, matching the existing claude-opus-4-7 Bedrock pattern. Only the Vertex entry includes tiered cache pricing, matching publishers/anthropic/models/claude-opus-4-7.

Resolver failure root causes addressed

1. Title mismatch: The title [BOT ISSUE] Bedrock/Vertex: add missing claude-opus-4-8 provider variants does not match any of the resolver's three title regex patterns (single-model, multi-model parenthesized, multi-model colon). The metadata block must supply provider and models since the title cannot be parsed.
2. Invalid provider: "provider": "bedrock,vertex" is rejected by normalizeProvider() which expects a single provider string. Fixed to "bedrock" (majority provider; 3 of 4 models are Bedrock).
3. Missing model_specs: The original metadata had no model_specs or model_spec, so the resolver could not determine format, pricing, or token limits. Added complete model_specs map.
4. Missing locations for Vertex model: requiresExplicitLocations() returns true for models starting with publishers/. Added "locations": ["global"] based on Anthropic Vertex documentation which shows global as the recommended and default endpoint.

sync_models vs proposed update

sync_models cross-check found differences. Official provider verification was used for the applied values, and sync_models discrepancies are listed below for review.

Model	Field	Proposed update	sync_models	sync_models source models
publishers/anthropic/models/claude-opus-4-8	catalog entry	present	missing	None

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
ai-proxy	Ready	Preview, Comment	May 29, 2026 9:23am

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 2aa1c8a2f0

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "Codex (@codex) review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "Codex (@codex) address that feedback".

[chatgpt-codex-connector]

Respect configured Vertex locations

For Vertex Anthropic requests, resolveVertexLocation gives modelSpec.locations precedence over the secret metadata location, so adding locations: ["global"] here forces every Claude Opus 4.8 Vertex call to /locations/global/. In environments configured with a regional or multi-region Vertex location such as us-central1, us, or eu for residency/availability, this new model silently ignores that configuration, unlike the existing Anthropic Vertex models and the existing test coverage that expects metadata locations to be honored when model locations are absent.

Useful? React with 👍 / 👎.