FIELD NOTES / 5 MIN READ

Claude Opus 5.5 or Sonnet 5.5: Which Should Call Your Bot's Tools?

Anthropic's current lineup puts Claude Opus 5.5 and Claude Sonnet 5.5 side by side, both built for agentic tool use, at different prices and default effort levels. This note compares their official specs, flags the forced tool_choice breaking change both tiers share, and gives a rule for picking one for a small-team bot's tool-calling routine.

AI-assisted editorial: researched and drafted with AI using the primary sources linked below. Check the cited evidence and any stated test scope. Illustrative examples are not measured product results. Editorial policy.

Key takeaways

  • Sonnet 5.5 is Anthropic's own 'faster, lower-cost complement' to Opus 5.5, built for well-scoped coding, agent, and tool-use workloads — not a downgrade for tool calling specifically.
  • Both current tiers reject tool_choice of type any or tool with a 400 error. If your bot forces a tool call, you must switch to auto plus a strict schema before upgrading.
  • Anthropic's own guidance is to build an eval set on your actual prompts and data before choosing a tier — not to default on name or price alone.

The short answer for a tool-calling bot

If your bot's routine calls one or two tools for a clearly scoped task — check a calendar, fetch a record, draft a reply — start with Claude Sonnet 5.5. Anthropic's own platform docs describe it as offering the best combination of speed and intelligence, built for everyday coding, agent, and enterprise workloads including agentic tool use. Anthropic separately describes Sonnet 5.5 as a faster, lower-cost complement to Opus 5.5 that excels at well-scoped everyday tasks across coding, agents, and knowledge work.

Reserve Claude Opus 5.5 for routines where the agent has to chain many tool calls with judgment in between — deciding which tool to try next, recovering from a dead end, or synthesizing results across a long multi-step job. Anthropic's general guidance for teams who aren't sure which model to use is to start with Opus 5.5 for most workloads, which is a broader recommendation than the tool-calling case specifically. For a bot whose job is narrow and repeatable, that general default doesn't automatically win.

Sources: Claude Sonnet 5.5 - Claude Platform Docs, Claude Sonnet 5.5, Choosing the right model - Claude Platform Docs.

What actually differs between the two tiers

Both models are current and both support tool use: Anthropic's models overview states that all current models support text and image input, text output, multilingual capabilities, vision, and tool use. The split shows up in price, latency, and default reasoning depth, not in whether tool calling works at all.

Per Anthropic's own comparison table: Claude Opus 5.5 runs at $4 input / $20 output per million tokens, with moderate comparative latency, adaptive thinking always on, and a medium default effort level. Claude Sonnet 5.5 runs at $2 input / $10 output per million tokens, with fast comparative latency and a high default effort level. Both carry a 1M-token context window and a 128K-token max output, and both have the same June 2026 knowledge cutoff.

Sonnet 5.5's own model page confirms the pricing and positioning directly: it offers the best combination of speed and intelligence and is built for everyday coding, agent, and enterprise workloads, such as code generation, data analysis, content creation, visual understanding, and agentic tool use. On a straight per-token basis, Sonnet costs half of Opus on both input and output.

Sources: Models overview - Claude Platform Docs, Claude Opus 5.5 - Claude Platform Docs, Claude Sonnet 5.5 - Claude Platform Docs.

The failure path: forced tool_choice breaks on both tiers

If your bot currently forces a specific tool call — tool_choice of type any or tool — check this before moving to either current tier. Anthropic's Sonnet 5.5 migration guide is explicit: every earlier model accepted tool_choice of type any or tool, but Claude Sonnet 5.5 rejects both with a 400 error, including on the token counting endpoint. The fix the guide gives is to send tool_choice of type auto and mark the tool strict so its input matches the schema.

Opus 5.5 carries the same restriction. Its model page lists forced tool use returning an error as one of four breaking changes affecting code that ran on Claude Opus 5, alongside thinking that can no longer be disabled, thinking blocks now tied to the model and conversation that produced them, and the older computer_20251124 computer-use tool no longer being accepted on the Claude API and Claude Models on Google Cloud.

There's a quieter change worth testing for before you ship: Opus 5.5's docs note that text between tool calls now comes back inside thinking blocks, whose text is empty at the default display setting. An application that streams that in-between text to users will go quiet during multi-tool runs until the display setting is changed. If your bot's UI shows live status between tool calls, this is the kind of thing that passes a smoke test and then goes silent in production.

Sources: Claude Sonnet 5.5 migration guide - Claude Platform Docs, Claude Opus 5.5 - Claude Platform Docs.

Illustrative example: pricing a daily routine

This is a hypothetical calculation to show the mechanics, not a measured BotBento test. Say a bot runs a daily routine that reads an inbox summary, calls a calendar tool to check for conflicts, and drafts one reply — roughly 8,000 input tokens and 1,500 output tokens per run, once a day.

At Opus 5.5's listed $4/$20 per million tokens, that run costs about $0.032 input plus $0.030 output, or roughly $0.062 per day. At Sonnet 5.5's listed $2/$10 per million tokens, the same token counts cost about $0.016 input plus $0.015 output, or roughly $0.031 per day — half the price, before any difference in how many tokens each model actually uses to finish the task. Neither figure accounts for prompt caching, batch discounts, or the fact that a higher-effort model may use more or fewer tokens to reach the same result; those depend on your actual prompts and are not something to assume from the price sheet alone.

The question this example can't answer is whether Sonnet 5.5 finishes the same task correctly as often as Opus 5.5 on your specific tools and prompts. That has to come from your own evaluation, not from either model's list price.

Sources: Claude Opus 5.5 - Claude Platform Docs, Claude Sonnet 5.5 - Claude Platform Docs.

The decision rule

Anthropic's own choosing-a-model guidance applies regardless of which two tiers you're comparing: create benchmark tests specific to your use case, test with your actual prompts and data, and weigh performance and cost tradeoffs rather than assuming the more expensive tier wins. Having a good evaluation set is described as the most important step in that process.

For a small-team bot, that translates into a concrete check: run your real tool-calling routine — the same tools, the same prompts, the same task — on both Sonnet 5.5 and Opus 5.5, and compare pass rate, token spend, and latency side by side. If Sonnet 5.5 completes the task reliably, its lower per-token price and faster comparative latency make it the better default for that routine. Only move to Opus 5.5 for the specific routines where your own evals show Sonnet falling short — long chains of tool calls, ambiguous intermediate decisions, or tasks where a wrong tool choice is expensive to recover from. Anthropic also documents a multi-model pattern worth considering instead of an either/or choice: an orchestrator on the more capable tier delegates bulk work to a cheaper worker model, so most tokens bill at the lower rate while judgment calls still go to the stronger model.

Either way, re-check the tool_choice behavior and the thinking-block display change before you switch a production bot's model ID. A tier swap on Anthropic's current lineup is not a drop-in replacement for code written against last generation's models.

Sources: Choosing the right model - Claude Platform Docs.

Primary sources

Sources checked 2026-11-08. Standards and product documentation can change; follow the linked version when implementing.

  1. Choosing the right model - Claude Platform Docs — Anthropic
  2. Claude Opus 5.5 - Claude Platform Docs — Anthropic
  3. Claude Sonnet 5.5 - Claude Platform Docs — Anthropic
  4. Claude Sonnet 5.5 migration guide - Claude Platform Docs — Anthropic
  5. Claude Sonnet 5.5 — Anthropic
  6. Models overview - Claude Platform Docs — Anthropic

BotBento is in development. Suggest a correction.