fix: update Anthropic model metadata for claude-opus-4-8

[github-actions]

fix: update Anthropic model metadata for claude-opus-4-8

Closes #672

Source issue: #672

Summary

Field	Value
Provider	anthropic
Primary model	claude-opus-4-8
Changed models	claude-opus-4-8
Added models	None
Updated models	claude-opus-4-8
Verification sources	1 2

Verified metadata

Model	Display name	Parent	Providers	Format	Flavor	Token limits	Pricing	Lifecycle
claude-opus-4-8	Claude 4.8 Opus		anthropic	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

Verification notes

Verification

Official source verification

1. https://platform.claude.com/docs/en/docs/about-claude/models/all-models

   • Verified: claude-opus-4-8 is the current <NextOpus /> model
   • Verified: "Adaptive thinking: Yes" — confirms reasoning_budget: true
   • Verified: Context window 1M tokens — confirms max_input_tokens: 1000000
   • Verified: Max output 128k tokens — confirms max_output_tokens: 128000
   • Verified: Pricing $5/MTok input, $25/MTok output — confirms input_cost_per_mil_tokens: 5, output_cost_per_mil_tokens: 25

2. https://platform.claude.com/docs/en/docs/about-claude/pricing

   • Verified: <NextOpus /> row shows 5m Cache Writes = $6.25/MTok — confirms input_cache_write_5m_cost_per_mil_tokens: 6.25
   • Verified: <NextOpus /> row shows 1h Cache Writes = $10/MTok — confirms input_cache_write_1h_cost_per_mil_tokens: 10
   • Verified: Cache Hits & Refreshes = $0.50/MTok — confirms input_cache_read_cost_per_mil_tokens: 0.5

sync_models (LiteLLM) cross-check

Compared against anthropic.claude-opus-4-8 in https://raw.githubusercontent.com/BerriAI/litellm/refs/heads/main/litellm/model_prices_and_context_window_backup.json (litellm_provider: bedrock_converse):

Field	Proposed value	sync_models value	Match
max_input_tokens	1000000	1000000	Yes
max_output_tokens	128000	128000	Yes
input_cost_per_mil_tokens	5	5 (from 5e-06/token)	Yes
output_cost_per_mil_tokens	25	25 (from 2.5e-05/token)	Yes
input_cache_read_cost_per_mil_tokens	0.5	0.5 (from 5e-07/token)	Yes
input_cache_write_cost_per_mil_tokens	6.25	6.25 (from cache_creation_input_token_cost 6.25e-06)	Yes
input_cache_write_5m_cost_per_mil_tokens	6.25	6.25 (same as cache_creation_input_token_cost)	Yes
input_cache_write_1h_cost_per_mil_tokens	10	10 (from cache_creation_input_token_cost_above_1hr 1e-05)	Yes
reasoning_budget	true	No direct equivalent (supports_reasoning=true, supports_xhigh_reasoning_effort=true present)	N/A — verified from official Anthropic docs instead

No deviations found between proposed values and sync_models. All numeric fields match.

Fields not published or not applicable

• deprecation_date: Not applicable — claude-opus-4-8 is the current flagship model, not deprecated.
• parent: Not applicable — claude-opus-4-8 is the latest Opus model; it is not a dated snapshot of another alias.
• supported_regions: Not applicable — available_providers is ["anthropic"] only, not vertex.

sync_models vs proposed update

sync_models cross-check found no pricing/token discrepancies for the changed models.

No sync_models discrepancies were noted.

[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 11:14am

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: eea04dffe3

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]

Include the missing cache write tiers

This update only adds reasoning_budget, but the claude-opus-4-8 metadata still lacks the tier-specific input_cache_write_5m_cost_per_mil_tokens: 6.25 and input_cache_write_1h_cost_per_mil_tokens: 10 fields that the source issue says should be present. For Anthropic prompt-caching usage, especially 1-hour cache writes, consumers of this model metadata still cannot price the correct tier and may fall back to the generic 5-minute write cost or treat it as unknown, so the metadata remains incomplete after this fix.

Useful? React with 👍 / 👎.