opendunes
Evaluate and optimise
Copy page ▾

Move to Opus 5.5

Move an app from Claude Opus 5 to Claude Opus 5.5: the model IDModel IDThe name a request calls a model by, lab/model, copied from Models.More on Models → changes, and three settings with it.

  1. Change the model ID.
  2. Update the request.
  3. Read the answer.
  4. Set the effort.
  5. Compare before you switch.

1. Change the model ID

FromTo
Model IDanthropic/claude-opus-5anthropic/claude-opus-5.5
Context1 000 000 tokens1 000 000 tokens
Longest answer128 000 tokens128 000 tokens
Reasoningon unless you turn it offalways on: reasoning.mandatory is true
Effort levelslow · medium · high · xhigh · maxthe same five
A forced tool calltakenrefused

GET /v1/model/{lab}/{model} answers each model's entry: its context, its longest answer, its reasoning. The request keeps its shape; model changes.

JSON
{
  "model": "anthropic/claude-opus-5.5",
  "reasoning": {
    "effort": "medium"
  },
  "messages": [
    {
      "role": "user",
      "content": "Find the bug in this function and propose a fix."
    }
  ]
}

2. Update the request

You send todayOn Opus 5.5Send instead
reasoning: {"enabled": false}, effort none, or thinking: {"type": "disabled"}refused: reasoning cannot be turned offa lower effort
a thinking budget: reasoning.max_tokens or thinking.budget_tokensnot honouredan effort level
tool_choice set to required or to one tool by name; any or tool on /v1/messagesrefusedauto, with the prompt naming the tool to use

auto lets the model decide, so check that a call came back and ask again if not: see Tools and function calling. Where a forced call only served to get JSON back, ask for it with Structured outputs.

Don't

Edit an earlier turn, the system prompt or the tool list mid-conversation. Opus 5.5 ties each block of reasoning to the conversation before it, and an edit breaks the tie.

Do

Add to the conversation instead: a new instruction goes in a new message after the last one.

3. Read the answer

Every answer now carries reasoning, and the reasoning shares the ceiling with the reply.

RouteWhere the reasoning landsChange
/v1/chat/completionschoices[0].message.reasoning, counted in usage.completion_tokens_details.reasoning_tokensraise max_tokens: too low a ceiling ends with no text
/v1/messagesthinking blocks, ahead of the text blockread blocks by type, never content[0]; send thinking blocks back unchanged with tool results

Between tool calls, Opus 5.5 writes its progress notes into thinking blocks, where Opus 5 wrote them as text. On /v1/messages, thinking.display picks what those blocks hold: summarized for a summary of the reasoning, omitted for nothing.

JSON
{
  "model": "anthropic/claude-opus-5.5",
  "max_tokens": 64000,
  "thinking": {
    "type": "adaptive",
    "display": "summarized"
  },
  "output_config": {
    "effort": "medium"
  },
  "messages": [
    {
      "role": "user",
      "content": "Review the open changes to the billing module."
    }
  ]
}

Both routes: Chat · Messages.

4. Set the effort

Effort is the one control left over how long Opus 5.5 thinks, and so over what an answer costs.

RouteField
/v1/chat/completionsreasoning.effort, or reasoning_effort
/v1/messagesoutput_config.effort

Anthropic gives medium as Opus 5.5's default, one level under Opus 5's high. Set the level in every request rather than carry over the one Opus 5 ran at: start at medium, lower it where answers hold, raise it where they fall short.

Good to know

At xhigh or max, Anthropic's advice is a max_tokens of 64 000 to start: the reasoning and the reply both fit.

5. Compare before you switch

Send the same set of real prompts to both IDs, and compare three things.

CompareWhere
the answersyour own check: a test suite, a reviewer, the task's pass rate
what each costusage.cost, in DADAThe Algerian dinar, the one currency the docs show.More on Balance and charges →, on every answer
how much each reasonedusage.completion_tokens_details.reasoning_tokens

The script sends one task to the old ID, then to the new ID at three effort levels, then to the new ID with reasoning turned off. It prints one line per run: how the answer ended, what the model reasoned, and the charge in DA, then the first words of the answer. A refused request prints its status. Effort is the one control left over what an answer costs, so the three levels are the comparison to read.

import os
from openai import APIStatusError, OpenAI

API_KEY = os.environ["OPENDUNES_API_KEY"]
BASE_URL = "https://opendunes.com/v1"
MODEL = "anthropic/claude-sonnet-4.5"
CANDIDATE = "anthropic/claude-sonnet-4.5"
EFFORTS = ["medium", "high", "xhigh"]
MAX_TOKENS = 64000
TASK = (
    "Find the bug in this function and propose a fix.\n\n"
    "def average(values):\n"
    "    return sum(values) / len(values)"
)
OFF = {"reasoning": {"enabled": False}}
RUNS = (
    [("old", MODEL, {})]
    + [
        (effort, CANDIDATE, {"reasoning": {"effort": effort}})
        for effort in EFFORTS
    ]
    + [("off", CANDIDATE, OFF)]
)

client = OpenAI(base_url=BASE_URL, api_key=API_KEY)

for label, model, fields in RUNS:
    try:
        reply = client.chat.completions.create(
            model=model,
            max_tokens=MAX_TOKENS,
            messages=[{"role": "user", "content": TASK}],
            extra_body=fields,
        )
    except APIStatusError as error:
        print(f"{label:7} refused, status {error.status_code}")
        continue
    usage = reply.usage
    details = usage.completion_tokens_details
    thought = details.reasoning_tokens if details else 0
    print(
        f"{label:7} {reply.choices[0].finish_reason:7}",
        f"thought {thought}",
        f"cost {usage.cost}",
    )
    print("   ", (reply.choices[0].message.content or "")[:60])

The samples put anthropic/claude-sonnet-4.5 in both seats, so they run as they stand, and the off run is taken. Set MODEL to anthropic/claude-opus-5 and CANDIDATE to anthropic/claude-opus-5.5: the off run is then refused, because reasoning cannot be turned off.

Start at medium, Anthropic's default for Opus 5.5, and move up only where high or xhigh answers better on your own check. max_tokens is set to 64 000, Anthropic's advice at xhigh, so the reasoning and the reply both fit.

Switch once the new ID holds; going back is the same one-line change. Run your red-team suite on the new ID too: Red teaming.

Last modified on