Claude Opus 4.5: What Changed for Coding and Agents
Anthropic released Claude Opus 4.5 in November 2025 at $5 and $25 per million tokens, with an effort setting that trades depth for tokens, stronger resistance to prompt injection and state-of-the-art coding scores.
Tecyfy · Published · Updated · 8 min read
Written in December 2025, when this was news. Models and prices have moved on since: today's are on Model prices.
Anthropic released Claude Opus 4.5 on 24 November 2025 and called it the best model in the world for coding, agents and computer use. Two things changed for people paying for it: Opus became much cheaper, and developers got a dial for how hard the model works. This article explains both, as they stood at release.
A third of the price
Opus 4.5 cost $5 per million input tokensTokenThe small piece of text a language model reads and writes, often part of a word. Prices, limits and the context window are all counted in tokens. and $25 per million output tokens through the Claude API, a third of the earlier Opus price. Anthropic said the price made Opus-level capabilities accessible to more users, teams and enterprises. It was available in the Claude apps, the API and all 3 major cloud platforms, under the id claude-opus-4-5-20251101.
The effort parameter
Opus 4.5 introduced an effort setting on the API. Low effort answers quickly with fewer tokens; high effort lets the model keep working on a problem. Anthropic's own figures:
- At medium effort, Opus 4.5 matched Sonnet 4.5's best score on SWE-bench Verified while using 76% fewer output tokens.
- At high effort it beat Sonnet 4.5 by 4.3 percentage points while using 48% fewer tokens.
Because output tokens are what you pay most for, effort is a cost control as much as a quality one. See how contextContext WindowEverything a model can see at once while it writes an answer, counted in tokens. When the window is full, something has to be left out. and output size drive cost on Model prices.
Coding and agents
Anthropic reported state-of-the-art results on real-world software engineering tests, and said Opus 4.5 scored higher than any human candidate on its own 2-hour engineering take-home exam. It was built for long agenticAgentic WorkflowA fixed sequence of model calls and tools that your code controls, such as classify, then look up, then draft. An agent decides its own next steps; a workflow follows the path you wrote. work: with effort control, context compaction and better tool useTool UseGiving a model a list of things it can ask to have done, such as search the web or read a file. The model asks; the program around it does the work and hands back the result., it could run longer with less help.
A tool is described to the model as a name, a description and a JSON schema for its input. The model decides when to call it:
import anthropic
client = anthropic.Anthropic() # reads ANTHROPIC_API_KEY
tools = [{
"name": "read_source_code",
"description": "Read a file from the repository.",
"input_schema": {
"type": "object",
"properties": {"file_path": {"type": "string", "description": "Path from the repository root"}},
"required": ["file_path"],
},
}]
response = client.messages.create(
model="claude-opus-4-5-20251101",
max_tokens=4096,
system="You are a security reviewer. Look for SQL injection and XSS.",
tools=tools,
messages=[{"role": "user", "content": "Start with api/v1/users.ts."}],
)
print(response.stop_reason) # "tool_use" when the model asks for a file
Your code runs the tool and sends the result back in the next message; that loop is what makes an AgentAgentA model that uses tools in a loop, deciding each next step from what the last one returned, until the task is done.. The AI agents page follows one request through it step by step.
Harder to trick
With Opus 4.5 Anthropic reported large gains against prompt injectionPrompt InjectionAn attack that hides instructions in text a model reads, such as a web page, an email or a file, so the model follows the attacker instead of you., where instructions hidden in a web page or file try to steer the model. Anthropic said it was harder to trick with prompt injection than any other frontier model in the industry at the time.
In the apps
The same release changed the Claude apps: long conversations no longer stopped at a limit, because Claude summarised earlier parts as needed. Claude for Chrome reached all Max users, and Claude for Excel reached Max, Team and Enterprise users in beta.
What it means now
Opus 4.5 was a record of a turn in pricing: the top model became accessible to more users, teams and enterprises, with effort as a way to spend fewer tokens. Newer Claude models have followed since; the current ones and their prices are on Model prices.
Questions
- When was Claude Opus 4.5 released?
- Anthropic released Claude Opus 4.5 on 24 November 2025, under the API id claude-opus-4-5-20251101. It was available in the Claude apps, the API and all 3 major cloud platforms.
- How much does Claude Opus 4.5 cost?
- At release Opus 4.5 cost $5 per million input tokens and $25 per million output tokens through the Claude API. Current prices are on Tecyfy's Model prices page.
- What is the effort parameter in Claude Opus 4.5?
- Effort is an API setting for how hard the model works on a request. Low effort answers quickly with fewer tokens; high effort lets it keep working. Anthropic reported that at medium effort Opus 4.5 matched Sonnet 4.5's best SWE-bench Verified score with 76% fewer output tokens.
- Is Claude Opus 4.5 resistant to prompt injection?
- Anthropic reported large gains against prompt injection with Opus 4.5. Anthropic said it was harder to trick with prompt injection than any other frontier model in the industry at the time.
Sources