Anthropic Cuts Agent Costs with New Advisor Tool

Anthropic Cuts Agent Costs with New Advisor Tool

Ondřej Barták
Ondřej Barták
Entrepreneur and Programmer
13. 4. 2026
2 minutes reading · 23 views
Anthropic Cuts Agent Costs with New Advisor Tool

    Developers working with AI agents keep facing the same conundrum: you want a smart model, but you do not want to pay for every token from the top-tier Opus. What can you do? Anthropic has come up with an answer that part of the community had effectively invented on its own. They call it the advisor strategy.

    The principle is surprisingly simple. A smaller and cheaper model (Sonnet or Haiku) manages and executes the entire task. It calls tools, reads the results, and keeps working. But as soon as it encounters a problem it cannot solve on its own, it calls Opus. Opus reviews the context, gives it a plan or correction, and Sonnet continues. Opus itself never calls tools or writes outputs for the user. It only provides advice.

    Benefits and how to use it

    The test results are solid. Sonnet with Opus as an advisor scored 2.7 percentage points higher on the SWE-bench Multilingual benchmark than Sonnet alone. At the same time, the cost per task fell by almost 12%. That may sound modest, but it adds up very quickly at higher task volumes.

    The results with Haiku are even more interesting. On the BrowseComp benchmark, Haiku with Opus as an advisor achieved a score of 41.2%, while on its own it managed only 19.7%. It still lags behind Sonnet (which scores approximately 70% on its own), but costs 85% less per task. For companies processing thousands of requests per day, this is a very important figure.

    Anthropic has packaged the entire approach into a tool called the Advisor tool, which is available through the Messages API. All communication between the models takes place within a single API request. Sonnet simply calls the advisor, receives a response, and continues. The setup is simple:

    response = client.messages.create(
        model="claude-sonnet-4-6",
        tools=[
            {
                "type": "advisor_20260301",
                "name": "advisor",
                "model": "claude-opus-4-6",
                "max_uses": 3,
            },
        ],
        messages=[...]
    )

    The max_uses parameter limits how many times Sonnet may consult the advisor within a single request. Tokens from both models are billed separately at their respective rates, so you retain a clear overview of the costs.

    Who does it make sense for?

    The Advisor tool is currently in beta and works only on Anthropic's platform. It can be added with one extra line of code and a single header in the API request. Anthropic recommends testing three configurations side by side: Sonnet alone, Sonnet with Opus as an advisor, and Opus alone, so you can see exactly where the sweet spot lies for your use case.

    Companies such as Genspark and Eve Legal have already confirmed the results. Eve Legal reports that Haiku with Opus as an advisor delivers top-tier model quality on document extraction tasks, but at one-fifth of the cost. It is hard to find a reason not to at least give this a try.

    Advertisement

    Content created with help from UpTier.

    SEO and GEO on autopilot. UpTier’s multi-agent systems write and optimize content for search engines and AI answers.

    Discover UpTier ↗

    Category:AI
    Did you enjoy this article?
    Discover more interesting posts on our blog
    Back to blog

    Related posts

    OpenAI introduces GPT-6.2 Astra and Sol for professional and everyday workOpenAI introduces GPT-6.2 Astra and Sol for professional and everyday work
    OpenAI has introduced two models: Astra for complex professional tasks and Sol for everyday work. Both are available in ChatGPT Work, Codex and through the API.
    3 min read
    17. 11. 2026
    Tencent releases Hy4 Flash with image support and a context window of up to 1 million tokensTencent releases Hy4 Flash with image support and a context window of up to 1 million tokens
    The lighter Hy4 Flash model targets chat, documents and routine code edits. It accepts text and images, supports tool use and, according to Tencent, matched Claude Haiku 5.5 in tests of everyday tasks.
    1 min read
    16. 11. 2026
    Companies Linked to Chinese Spy Invested in DeepSeek and Promised Banned Nvidia ChipsCompanies Linked to Chinese Spy Invested in DeepSeek and Promised Banned Nvidia Chips
    Companies linked to an alleged Chinese spy joined DeepSeek’s latest funding round. In exchange for a stake, they reportedly promised banned Nvidia Blackwell chips.
    3 min read
    16. 11. 2026
    Přihlaste se k odběru našeho newsletteru
    Zůstaňte informováni o nejnovějších příspěvcích, exkluzivních nabídkách, a aktualizacích.
    CodedTrip

    Operated by CodedTrip LLC, USA.

    YouTube
    TikTok