Claude Opus 4.1 was positioned for real-world coding and reasoning-heavy work, including debugging and multi-file changes. But as of November 12, 2026, its direct Anthropic API model ID, claude-opus-4-1-20250805, is retired. Anthropic says requests to retired models fail, so do not use that ID for a new direct-API integration. If you are evaluating Opus 4.1 through another provider, check that provider’s separate lifecycle and availability information. (Anthropic model deprecations)
Was Claude Opus 4.1 good for coding?
Anthropic positioned Opus 4.1 as an upgrade for agentic tasks, real-world coding, and reasoning, and reported 74.5% on SWE-bench Verified. Anthropic’s announcement says the SWE-bench result used no extended thinking; its Claude 4 benchmark scaffold used a bash tool and a file-editing tool. That is a result under a stated benchmark setup, not a prediction of success on your repository. (Anthropic Opus 4.1 announcement)
For coding-model selection, that makes Opus 4.1 relevant as a historical benchmark and a candidate to evaluate where it remains available. Examples worth testing include difficult debugging, planning a multi-file refactor, and reviewing a complex codebase. A benchmark score does not show that it will outperform a current model on those tasks.
A third-party comparison recommends Sonnet 4 for routine, high-volume, or cost-sensitive work and Opus 4.1 for more complex reasoning. It reports SWE-bench Verified scores of 72.7% for Sonnet 4 and 74.5% for Opus 4.1, but the captured article does not provide test methodology. Treat that comparison as reported context, not proof of which model will work better for your code. Developer anecdotes about relative speed and quality are mixed, too. (Adaline Labs comparison)
For a new model choice, compare versions that are available on your intended API channel. Anthropic’s current Messages API documentation demonstrates a request with claude-opus-5-5; confirm model availability for your account before deployment. A workflow that uses one model to plan and another to implement is a suggestion to test, not a guaranteed feature or cost saving. (Anthropic Messages API guide)
Direct Anthropic API: status and example
Do not send claude-opus-4-1-20250805 to the direct Anthropic API: Anthropic’s lifecycle documentation lists that model as retired on August 5, 2026. Retirement schedules on partner-operated platforms can differ, so this status should not be assumed to describe Bedrock or other providers. (Anthropic model deprecations)
The example below shows the direct Anthropic Messages API using claude-opus-5-5 as in Anthropic’s current SDK guide. It is not a way to call Opus 4.1. Set ANTHROPIC_API_KEY to a Console API key and keep the key out of source control. The Anthropic SDK handles authentication, the API version, and content-type headers. A Messages request includes a model, max_tokens, and a messages array; the returned message’s content contains blocks, including text blocks. (Anthropic API overview; authentication; Python SDK)
import os
from anthropic import Anthropic
client = Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])
message = client.messages.create(
model="claude-opus-5-5",
max_tokens=512,
messages=[
{
"role": "user",
"content": "Suggest edge cases for a function that retries failed HTTP requests. Return test ideas only.",
}
],
)
for block in message.content:
if block.type == "text":
print(block.text)
If you make raw HTTP requests instead of using an SDK, use the Messages endpoint and the documented API-version header; Anthropic currently recommends using the latest API version when possible. (Anthropic API versioning; Messages API reference)
Opus 4.1 API pricing: historical rates and an estimate
Anthropic’s pricing page lists Opus 4.1 at $15 per million input tokens and $75 per million output tokens. Because the direct API model is retired, treat these as historical listed rates—not a quote for a request you can make to Opus 4.1 on the direct API today. They are API token rates, not a monthly Claude subscription price. (Anthropic pricing; model deprecations)
| Historical Opus 4.1 direct API rate | Price |
|---|---|
| Input | $15 per million tokens |
| Output | $75 per million tokens |
At those listed rates, 100,000 input tokens and 20,000 output tokens would calculate to $3 before any other billing adjustments:
- Input: 0.1 million × $15 = $1.50
- Output: 0.02 million × $75 = $1.50
- Total: $3.00
This is an arithmetic illustration of the listed rates, not a current Opus 4.1 API cost estimate. For a replacement model, use its current rates and actual metered token use. Anthropic’s pricing page lists Opus 5.5 at $4 per million input tokens and $20 per million output tokens; a lower per-token price alone does not establish lower cost per completed coding task. (Anthropic pricing)
How to evaluate a coding-model replacement
Compare a currently available model with the alternative you are considering on representative tasks. Use the same task description, repository context, test commands, and acceptance criteria. Record:
- Task outcome: Did the change meet requirements and pass relevant tests?
- Corrections: How many follow-up prompts or code changes were needed?
- Human effort: How much time went to review, debugging, and rework?
- Elapsed time: How long did it take to reach an acceptable result?
- Actual API cost: What did the provider meter for input and output tokens, plus any applicable billing categories?
Choose based on your observed results, not a benchmark score or a generic ROI threshold. When changing models, test the replacement on your own tasks before adopting it in production. (Anthropic model deprecations)
Sources
Model deprecations — Claude Platform Docs
Claude Sonnet 4 Vs Opus 4.1: Which Model To Use For Coding
Using the Messages API — Claude Platform Docs
API overview — Claude Platform Docs
Authentication — Claude Platform Docs
Python SDK — Claude Platform Docs
Versions — Claude Platform Docs