Fix the Make Claude Sonnet 5.5 Extended Thinking Error: Replace Disabled with Between Tools

Introduction

You changed one dropdown, the model name, and a scenario that ran for weeks suddenly fails. That is the Make and Claude Sonnet 5.5 extended thinking error: the scenario pairs Sonnet 5.5 with Extended Thinking set to Disabled, Make reports “Value not found in options”, and no API call ever reaches Anthropic. This tutorial is for Make users who run Anthropic Claude modules in production scenarios and need the scenario working again without redesigning the whole flow. You will learn why the old value is rejected, how to confirm the diagnosis in under a minute, and how to fix it two ways: inside the standard module where possible, and reliably through the Make an API Call module where you control the exact request body.

Quick answer

In Make, stop sending Disabled thinking to Claude Sonnet 5.5. Use the Anthropic Claude Make an API Call module with thinking set to between_tools and effort at high or below, because Sonnet 5.5 rejects disabled thinking.

Evidence: what actually changed in Sonnet 5.5

Anthropic launched Claude Sonnet 5.5 on September 28, 2026, across the Claude API and major cloud platforms, and described it as generating output more than 30 percent faster than Claude Sonnet 5 while keeping the 1-million-token context window and 128,000-token output limit. That speed claim matters here because it explains why teams switch models in the first place: the upgrade looks like a free performance win, so the model dropdown gets changed and the thinking setting is left exactly as it was. The thinking change is the trap. On Sonnet 5, turning thinking off meant sending thinking: {"type": "disabled"}. On Sonnet 5.5, that value is rejected, and the lowest thinking setting is between_tools instead.

Anthropic states the rule plainly in its thinking troubleshooting documentation: “On Claude Sonnet 5.5, send thinking: {type: \"between_tools\"} to turn off up-front thinking.” The same page adds the effort constraint that decides whether your fix works: Sonnet 5.5 accepts between_tools at effort high or below, and combining it with effort xhigh or max returns a 400 error. The extended thinking documentation describes adaptive thinking as the newer approach, where Claude decides whether and how deeply to think on each request, with depth steered by the effort parameter. Those two pages together define the whole fix: a different value, a capped effort level, and no extra thinking fields.

The Make side of the story comes from the community. A Make community thread from October 5, 2026, reported Sonnet 5.5 in the Create a Prompt module failing with “Value not found in options” when Extended Thinking was set to Disabled, and the thread records that the issue was confirmed on the vendor side. Make’s own Anthropic Claude integration listing confirms the two module names this tutorial uses, Create a Prompt and Make an API Call, so the workaround below uses real modules, not hypothetical ones. If you already build agents in Make, our guide to building next-gen Make AI Agents in the scenario builder covers the agent surface this fix plugs into, and our Make error handlers guide shows how to keep a failed module from taking down the whole scenario while you test the fix.

Prerequisites

• A Make account with a scenario that uses the Anthropic Claude app, and permission to edit that scenario.
• An Anthropic API key connected to Make, or the ability to add one as a custom connection.
• The scenario’s current model setting noted down before you change anything, so you can roll back to Sonnet 5 if the fix needs more testing.
• A test input that reproduces the failure, ideally a short prompt with no side effects, so you can run the fixed scenario without sending real messages to customers.
• Basic comfort reading JSON, because the reliable fix edits a request body directly.

Step-by-step guide

Step 1: Confirm the diagnosis before changing anything

Open the failing scenario and run it once with the existing settings. If the Anthropic Claude module fails with “Value not found in options” and the module is set to a Sonnet 5.5 model with Extended Thinking set to Disabled, the diagnosis is confirmed: the model rejects the thinking value, not your prompt, your key, or your connection. Do not rotate the API key or rebuild the connection first. Both are red herrings for this error, and rotating a working key creates a second problem while the first one is still there. Note the exact model string in the module, because Sonnet 5 and Sonnet 5.5 behave differently here and the fix depends on which one you actually selected.

Step 2: Try the in-module fix first

If your scenario uses the Create a Prompt module, open that module and look at its thinking or extended thinking setting. Remove the Disabled selection: either pick an enabled or adaptive style option if the module offers one, or clear the thinking setting entirely so the request goes out without an explicit disabled value. Save the module and run the scenario with your test input. If the run succeeds, you are done, and the rest of this tutorial is background. If the module still fails, or if it offers no usable alternative to Disabled, move to Step 3 instead of fighting the dropdown. The module’s option list can lag behind the model’s accepted values, and that lag is exactly what the API-call route sidesteps.

Step 3: Replace the module with Make an API Call

Add an Anthropic Claude Make an API Call module in place of the failing module, keeping the same position in the scenario so upstream data and downstream mappings stay intact. In the API call module, use your existing Anthropic connection, set the method to POST, and set the URL to https://api.anthropic.com/v1/messages. This module sends exactly the request body you write, so no dropdown can substitute an outdated value for you. Map the prompt fields from the old module into the new body’s messages array, and map the old module’s output references in downstream modules to the new module’s response fields after your first successful test run.

Step 4: Write a request body Sonnet 5.5 accepts

Set the headers to Content-Type: application/json, x-api-key with your Anthropic key, and anthropic-version: 2023-06-01. Then use a body like this, with your own prompt in place of the example text:

{
  "model": "claude-sonnet-5-5",
  "max_tokens": 4096,
  "thinking": {
    "type": "between_tools"
  },
  "output_config": {
    "effort": "high"
  },
  "messages": [
    {
      "role": "user",
      "content": "Summarize this changelog in three bullet points."
    }
  ]
}

Three details in this body are load-bearing. First, thinking.type is between_tools, the value Sonnet 5.5 accepts for turning off up-front thinking. Second, effort is high, inside the low, medium, high range that between_tools allows. Third, the thinking object contains nothing else: no display, no budget_tokens, and no block_binding, because any extra field next to between_tools is itself a 400 error. Keep temperature, top_p, and top_k out of the body as well. Sonnet 5.5 migration coverage reports non-default sampling values returning 400 errors on this model, and the defaults are what you want for a deterministic scenario step anyway.

Step 5: Handle the no-tools case explicitly

Before you celebrate a passing test, check whether your scenario actually uses tools. With between_tools and no tools in the request, the model answers without thinking first, because there is no gap between tool calls for thinking to happen in. That is correct behavior, not a bug, but it changes output quality on reasoning-heavy prompts. If your scenario makes a plain text call with no tools and the prompt needs real reasoning, drop the thinking field from the body instead and let adaptive thinking decide how deeply to think. If your scenario does use tools, keep between_tools: the model thinks briefly between calls, and those short updates come back as thinking blocks in the response.

Step 6: Test, remap outputs, and re-enable the schedule

Run the scenario with your test input and inspect the module output. A successful run returns a message with content blocks; if you see thinking blocks, pass them back unchanged in any follow-up request in the same conversation, because the API requires thinking blocks to round-trip intact. Then remap any downstream module that read the old module’s output, since field paths change when a module is replaced. Run the full scenario once more end to end, and only then re-enable the schedule if you paused it. If the scenario feeds other systems, our Make scenario deduplication guide is worth reading before the first scheduled run, because a fixed scenario replaying a backlog can otherwise create duplicate records.

Troubleshooting

The error persists after switching to Sonnet 5.5 with a new module

If a freshly added module still reports “Value not found in options”, the old module’s configuration is probably still in the execution path: a second Claude module in the scenario, a router branch you did not open, or a copy of the scenario still on a schedule. Search the scenario for every Anthropic Claude module, not just the one in the failing log line, and check each one’s model and thinking settings. Also re-open the module you edited and confirm the model field actually saved as Sonnet 5.5; a module that failed validation on save can silently keep its previous selection.

A 400 error mentions effort after the thinking fix

This is the xhigh and max trap. between_tools is accepted only at low, medium, or high effort. If your scenario sets effort to xhigh or max, or if a per-message effort differs from the effort already in effect for the conversation, the API returns a 400 even though the thinking type is now correct. Set output_config.effort to high and remove any per-message effort override. If the task genuinely needs xhigh or max reasoning, drop the thinking field and use adaptive thinking instead, which is the documented path for those effort levels.

A 400 error appears after adding budget_tokens or display

between_tools takes no other field. Adding budget_tokens to control thinking length, display to see thinking summaries, or block_binding to control block behavior each produce a 400 on their own. Remove all three from the thinking object. If you need visible reasoning summaries, that is a reason to use adaptive thinking with a display setting, not to bolt display onto between_tools.

A 400 error appears after setting temperature, top_p, or top_k

Sonnet 5.5 migration coverage reports non-default sampling values being rejected on this model. Remove temperature, top_p, and top_k from the request body and re-run. If your prompt depended on a high temperature for varied output, move that variation into the prompt wording itself, for example by asking for three distinct options, rather than into a sampling parameter the model no longer accepts.

Thinking blocks break the next turn, or progress notes vanish

Two related symptoms have the same cause: thinking blocks that are dropped, edited, or reordered between turns. The API requires thinking blocks to be passed back unchanged, including any redacted thinking blocks, or the protocol breaks and later turns misbehave. In Make, map the full content array from the previous response into the next request rather than extracting only the text blocks. If progress notes seem to vanish, check your output mapping first: a mapping that reads only the first text block will hide thinking blocks that are present in the raw response.

FAQ

Can Claude Sonnet 5.5 turn thinking fully off?

No. Sonnet 5.5 does not accept thinking type disabled. The lowest available setting is between_tools, which turns off up-front thinking but still writes short progress notes between tool calls, and those notes come back as thinking blocks.

Does between_tools work in a scenario that uses no tools?

It runs, but the model answers without thinking first, because there are no tool calls for thinking to happen between. For a reasoning task with no tools, omit the thinking field instead and let adaptive thinking decide, or keep the task on a model that still accepts disabled.

Which effort levels work with between_tools?

Low, medium, and high. Effort xhigh or max returns a 400 error when combined with between_tools, and a per-message effort that differs from the effort already in effect also returns a 400.

Is switching back to Claude Sonnet 5 a real fix?

It is a temporary workaround, not a fix. Sonnet 5 still accepts disabled thinking, so the old scenario runs again, but every future model switch reopens the same problem. The durable fix is to send a thinking value the current model accepts.

Why does Make show ‘Value not found in options’ instead of an Anthropic error?

The message is Make’s module validation surfacing a value the chosen model does not accept. Make community reports from October 2026 describe exactly this symptom when Sonnet 5.5 is paired with Extended Thinking set to Disabled in the Create a Prompt module.

Does between_tools need a beta header?

No. Anthropic’s migration notes describe between_tools as needing no beta header. It takes no other thinking field either, so do not add display, budget_tokens, or block_binding next to it.

Conclusion

The Make Claude Sonnet 5.5 extended thinking error looks like a broken scenario, but it is really a stale value: a thinking setting from the Sonnet 5 era that Sonnet 5.5 no longer accepts. Replace Disabled with between_tools, keep effort at high or below, keep the thinking object free of extra fields, and route through Make an API Call when the standard module’s dropdown cannot express the right value. Your scenario then runs on the current model instead of depending on the previous one. Next, read our guide to connecting external MCP servers to a Make AI agent if your fixed scenario is part of a larger agent, and keep the complete n8n automation guide bookmarked as the pillar reference for how these automation patterns generalize beyond Make.

Frequently asked questions