Claude Opus 5 API Pricing and Claude Code Setup Guide

The claude-opus-5 route is now available through this gateway. It is positioned for complex coding, long-running agents, deep reasoning, and high-value knowledge work. Its current gateway rate is 20x (CNY 8.00 per 1M tokens).

The naming boundary matters. claude-opus-5 is this gateway's route ID. As of July 25, 2026, Anthropic's official model overview lists current public models including claude-fable-5, claude-opus-4-8, claude-sonnet-5, and claude-haiku-4-5-20251001. It does not list an official Claude API model named claude-opus-5. This page documents gateway access and pricing; it is not an Anthropic launch announcement, and it does not transfer benchmarks, context limits, or prices from another Claude 5 model to this route.

Claude Opus 5 route at a glance

Item	Current gateway information
Model ID	claude-opus-5
Rate	20x
Reference price	CNY 8.00 per 1M tokens
Context window	Listed as 1M in the gateway catalog
Vision	Listed as supported in the gateway catalog
Good first evaluations	Complex coding, architecture planning, code review, long-running agents, rigorous reports
API formats	OpenAI-compatible Chat Completions and Claude Code-compatible configuration
Official naming status	No identically named model in Anthropic's current public catalog

These are gateway catalog settings, not official Anthropic specifications. The 1M context and vision labels also require separate testing before production use. Recheck the live pricing page and model guide, then run a small request.

How to call the Claude Opus 5 API

Use the regional Base URL recommended in the API endpoint guide. This minimal example uses the global endpoint:

curl https://gpt-agent.cc/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-opus-5",
    "messages": [
      {
        "role": "user",
        "content": "Review this refactoring plan and list architecture risks, validation steps, and rollback conditions."
      }
    ]
  }'

The OpenAI Python SDK only needs the Base URL, key, and model ID changed:

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://gpt-agent.cc/v1",
)

response = client.chat.completions.create(
    model="claude-opus-5",
    messages=[
        {
            "role": "user",
            "content": "Plan a phased migration for this repository and define acceptance criteria for each phase.",
        }
    ],
)

print(response.choices[0].message.content)

Start with a short text prompt and record the status code, request ID, latency, and billed usage. Test tool calls, long context, images, and structured output independently. A Claude-family route name is not evidence that every capability matches a particular official Anthropic model.

Using claude-opus-5 with Claude Code

Claude Code's official model configuration guide supports selecting a model through --model, ANTHROPIC_MODEL, or settings. When Claude Code connects to an LLM gateway, the model value must be a route that the gateway actually accepts. Start with this session-scoped configuration:

export ANTHROPIC_AUTH_TOKEN="YOUR_API_KEY"
export ANTHROPIC_BASE_URL="https://api.llm-token.cn"
export ANTHROPIC_MODEL="claude-opus-5"

claude --model claude-opus-5

Run a read-only task first, such as summarizing the repository structure or explaining a test file. Only enable file writes, shell commands, or deployment permissions after confirming the selected model, authentication, and output. See the full Claude Code setup guide for installation and gateway configuration.

Strong first evaluation tasks

• Architecture analysis and migration planning across several modules
• Risk-first review of a large diff
• Agent workflows that search, edit, test, and review over multiple steps
• Evidence-based root-cause analysis for complex incidents
• High-value technical proposals, reports, and release checklists

Tasks that usually do not need it by default

• Short questions, classification, extraction, and formatting
• High-volume, low-risk code completion
• Batch jobs that a lower-cost model already completes reliably
• Autonomous execution without acceptance criteria, permission boundaries, or rollback

A high-capability route should not become the default for every request. A practical system uses lower-cost models for retrieval, classification, and simple edits, then reserves claude-opus-5 for architecture decisions, difficult debugging, critical reviews, and long-running agent stages.

How Claude Opus 5 gateway pricing works

The platform baseline is approximately CNY 0.40 per 1M tokens at 1x. The current claude-opus-5 rate is 20x, making its reference price CNY 8.00 per 1M tokens.

Estimated usage = billed tokens / 1,000,000 * CNY 8.00

This converts the shared gateway multiplier into a budgeting reference. The account ledger, cache treatment, and live pricing page remain authoritative. Agent budgets should include tool results carried back into context, failed retries, context compaction, and final output, not only the user's first prompt.

Evaluate 20 to 50 representative tasks and record:

1. first-pass completion and acceptance rate;
2. input, output, and cache-related tokens;
3. time to first token and total latency;
4. tool failures, retries, and interrupted runs;
5. engineer rework time;
6. final cost per accepted task.

Claude Opus 5 vs Opus 4.8, Fable 5, and Sonnet 5

This is a procurement and evaluation map, not a benchmark ranking:

Route or official model	Good first workload	Verify before choosing
Gateway claude-opus-5	High-value coding, review, planning, and agent workflows	Live gateway price, route capabilities, and target client compatibility
Official claude-opus-4-8	Complex agentic coding and enterprise work	Official specifications, price, and regional availability
Official claude-fable-5	Long-running agents that need Anthropic's highest widely available capability	Refusal and fallback handling, official pricing, and classifier behavior
Official claude-sonnet-5	Daily coding and agents that prioritize speed and cost	Introductory pricing deadline and subsequent standard rate

Anthropic positions Opus 4.8 for complex agentic coding and enterprise work, Fable 5 as its most capable widely released model for long-running agents, and Sonnet 5 as the speed-intelligence balance. Do not infer the backend of the gateway's claude-opus-5 route from its name. Compare it using the same tasks, prompts, tools, permissions, and acceptance tests.

For a cost-sensitive comparison set, add Qwen 3.8 Max and Kimi K3. Model selection should optimize cost per accepted task, not only cost per million tokens.

Production agent checklist

1. Copy claude-opus-5 from the model guide; do not invent or substitute another ID.
2. Validate the API key, Base URL, model, status code, and response shape with a short prompt.
3. Start Claude Code with read-only work before granting write or shell permissions.
4. Set explicit boundaries for files, databases, payments, releases, and server operations.
5. Configure tool timeouts, retry limits, idempotency, and human takeover conditions.
6. Pin the evaluation repository, starting commit, dependencies, and acceptance tests.
7. Set per-request token limits, daily budget alerts, and a lower-cost fallback.
8. Recheck live pricing before launch instead of treating this snapshot as a permanent quote.

Frequently asked questions

What is the Claude Opus 5 model ID?

Use claude-opus-5 on this gateway. It is a gateway route ID. As of July 25, 2026, Anthropic's public model catalog does not list an identically named Claude API model.

How much does the Claude Opus 5 API cost?

The current gateway rate is 20x, or approximately CNY 8.00 per 1M tokens. The pricing page and account ledger are authoritative.

Can Claude Opus 5 run in Claude Code?

You can test it through this compatible gateway. Set ANTHROPIC_BASE_URL, ANTHROPIC_AUTH_TOKEN, and ANTHROPIC_MODEL=claude-opus-5, then validate with a read-only task.

Is Claude Opus 5 an official Anthropic model?

That claim is not supported by the current official catalog. This guide only confirms an identically named compatible route on this gateway.

Which tasks should test Claude Opus 5 first?

Start with complex code reviews, cross-module refactoring plans, difficult incident analysis, and long-running agent work that has explicit acceptance criteria. Use lower-cost models for simple high-volume tasks.

Where can I buy API access?

New users can get a multi-model API key from the buy page; existing key holders can use top-up. This is an independent compatible gateway, not an official Anthropic storefront.

Primary sources

• Anthropic model overview: current public models, official IDs, pricing, and selection guidance
• Anthropic model IDs and versioning: official model ID structure
• What's new in Claude Opus 4.8: the current official Opus positioning
• Claude Code model configuration: --model, ANTHROPIC_MODEL, and gateway custom-model behavior
• Gateway pricing: current claude-opus-5 route and multiplier