Opus 5 holds the flagship spot and Opus 4.8 is the fallback, while Opus 4.6 and 4.7 keep running in the codexkey lineup unchanged. Nobody is in a hurry to switch them off. Some traffic is tied to these revisions on purpose.
Why older revisions stay available
Some pipelines were written and validated against one specific revision, and any silent model swap under the hood risks a regression. Say the prompt, the tool schema and the response parsing are tuned to how Opus 4.6 behaves. Dropping in a newer revision without tests is then not an upgrade. It is an experiment on production.
codexkey does not force a migration. Both revisions are called by their own ids, just like the flagship.
Who pins to a specific revision, and why
- Teams with a frozen eval dataset. Do you have a golden set of cases and a pass-rate metric tied to one revision? Then switching models breaks comparability of the numbers over time. A pin keeps the baseline where it was.
- Regulatory and audit requirements. Sometimes you must show on paper which model version handled a given request, especially when the output ends up in reporting or a legally significant document.
- Long-running agent scenarios. The agent is built around a particular reasoning style and answer format. Swap the revision mid-flight and behavior shifts partway through a scenario that runs for hours or days.
- Gradual rollout. Part of the traffic stays on the older revision on purpose, as a control group, while the newer one is exercised on the rest of the flow.
How 4.6 differs from 4.7
Anthropic does not publish a line-by-line changelog between minor Opus revisions. Specific behavioral differences are better checked against your own cases than taken on faith from someone else's article, this one included. If you already have an eval dataset, run it on both revisions and compare the numbers. That beats any writeup.
How to call a specific revision
Only model changes. Everything else works the same as with any codexkey model:
curl https://api.codexkey.online/v1/messages \
-H "x-api-key: sk_cdx_..." \
-H "anthropic-version: 2023-06-01" \
-H "content-type: application/json" \
-d '{
"model": "claude-opus-4.6",
"max_tokens": 1024,
"messages": [{"role": "user", "content": "Run this case from the golden dataset"}]
}'
The same call against revision 4.7:
from openai import OpenAI
client = OpenAI(api_key="sk_cdx_...", base_url="https://api.codexkey.online/v1")
resp = client.chat.completions.create(
model="claude-opus-4.7",
messages=[{"role": "user", "content": "Run this case from the golden dataset"}],
)
print(resp.choices[0].message.content)
Both ids sit on the same endpoint, both API formats behave identically, and one key covers either.
What long-term pinning risks
A pin removes today's risk of a silent regression but adds another: sooner or later Anthropic itself can mark any revision as deprecated. If the model id is hardcoded, set up a recurring task to check the revision's status and keep a migration plan ready. Otherwise you will learn about a shutdown from production errors.
The price does not depend on the revision
On codexkey, 4.6, 4.7, 4.8 and 5 cost the same: $0.3 per million input tokens and $0.3 per million output tokens. An older revision neither saves money nor costs extra. The difference is in behavior and stability, not in the bill. You can estimate the total for your own load with the calculator on the site, and per-key spend is broken out separately in the dashboard.
Bottom line
Opus 4.6 and 4.7 stay in service for teams that need reproducibility: frozen evals, audits, long agent scenarios, gradual rollouts. Check behavioral differences between revisions against your own cases. Every Opus revision costs the same, so a pin settles a question of stability, not of saving money.