“`finishReason`: `MALFORMED_FUNCTION_CALL`” — nothing was counted, so nothing can be raised

"finishReason": "MALFORMED_FUNCTION_CALL"

No limit was reached. The model tried to call one of your tools and the call it produced was not valid, so generation ended there. That is a different kind of ending from every cap on this site: the definition above is about the validity of one call, not about a count, so there is no number in it for you to raise. It is reported as a value of finishReason — a field the reference marks Output only, carried in the response body rather than signalled as a request failure — which is why it so often surfaces as “the agent returned nothing” rather than as an error.

In ten seconds. Raising max_iterations, max_turns or any other ceiling cannot fix this and will not change the symptom. The run did not run out of anything.

Read candidates[0].finishReason, and read finishMessage next to it — the reference states that field is populated only when finishReason is set, so this is one of the cases it exists for. Then look at your tool schema, because the call that could not be built is a call against that schema.

What the reference says, quoted

From Generating content, in the FinishReason enum, introduced on that page as Defines the reason why the model stopped generating tokens.:

MALFORMED_FUNCTION_CALL The function call generated by the model is invalid.

A note on how much of this is quoted and how much is ours. The sentence above is the whole of what that reference page says about this enum value: the string occurs twice in the 1,232,466 bytes we fetched — once in the rendered enum table, once in the page’s own embedded data — and carries that one definition both times. Everything after this paragraph is our reading of the consequences, not the vendor’s text, and it is marked off from the quotes for that reason.

Why this one is the odd page on this site

Every other error with a page here is a ceiling. LangGraph counts supersteps; the OpenAI Agents SDK counts turns; Pydantic AI counts requests; AutoGen counts messages; Gemini itself, on the neighbouring page, counts consecutive tool calls. In every one of those, the run was cut off while still capable of continuing, and the honest question is whether the cap or the work was wrong.

Here the run was not cut off. It stopped because the next thing it produced could not be used. That inverts the debugging order completely. On a cap you ask “how many steps does this task need, and why does it need that many?”. On this you ask “what about the call I asked for is impossible to express?” — and the answer is in the shape of your tool declarations rather than in any configuration number. It is worth being blunt about the practical cost of mistaking one for the other: a team that reads this as a limit raises every cap it can find, pays for longer runs, and gets the identical empty response at the end of each one.

Reading the field properly

The reference defines the field, and its companion, like this:

finishReason enum (FinishReason) Optional. Output only. The reason why the model stopped generating tokens. If empty, the model has not stopped generating tokens. finishMessage string Optional. Output only. Details the reason why the model stopped generating tokens. This is populated only when finishReason is set.

Two things follow. An absent finishReason does not mean “finished normally” — by the reference’s own words it means generation has not stopped, which is the streaming case, and code that treats the two alike will report partial chunks as complete answers. And finishMessage is the only per-incident detail the API offers. On a value as terse as this one, that single string is the difference between an actionable report and a shrug, and most response-handling code never reads it.

The neighbours, and why the distinction is the whole diagnosis

Three values in that enum describe a tool interaction that did not work out, and they have three different causes:

MALFORMED_FUNCTION_CALL The function call generated by the model is invalid. UNEXPECTED_TOOL_CALL Model generated a tool call but no tools were enabled in the request. MALFORMED_RESPONSE Finished due to malformed response.

UNEXPECTED_TOOL_CALL points squarely at your request: tools were wanted and none were declared, which in an agent framework usually means the tools array was lost somewhere between your agent definition and the wire — a conditional that did not fire, a config key that did not survive a refactor. MALFORMED_RESPONSE is wider than tool calling and says only that the response itself was malformed. The value on this page is narrower than both: tools were declared, a call was attempted, and the attempt was not a valid one. Logging the enum value verbatim is therefore not a formality — collapsing these three into “tool error” in your own logs destroys the only distinction that tells you where to look.

What to do about it

Record the enum value verbatim, then finishMessage. Before anything else. These two strings are the entire evidence base, and both are discarded by the common pattern of logging only the response text.
Treat it as a schema problem first. An invalid call is a call measured against the declaration you supplied. Over-complex parameter shapes — deep nesting, free-form objects, unions, loosely-typed catch-all arguments — give a model more ways to produce something unusable than a flat set of named scalars does.
Simplify before you retry. A retry against the same schema is a bet that the next sample differs. Fewer parameters, flatter types and an explicit enumeration of allowed values change the odds instead of re-rolling them.
Do not raise any cap in response to this. It is the one ending on this site where that reflex is guaranteed to be wasted, because no count was involved.
Make “200 with no usable answer” a first-class case in your loop. Any agent framework over this API needs one branch for a successful response that carries nothing to act on. Without it, every value in this enum — this one, its two neighbours, and the counting one next door — presents identically as an agent that quietly produced nothing.

Provenance

One document, for this page: the Gemini API reference page Generating content on Google’s own developer site — 1,232,466 bytes, HTTP 200, no redirects, fetched 4 October 2026 for this page. The page carries its own date, Last updated 2026-09-23 UTC, and its own licence: “Except as otherwise noted, the content of this page is licensed under the Creative Commons Attribution 4.0 License”. Every string in a quote block above is from that fetch, and the enum definitions are reproduced word for word rather than paraphrased. We read that one reference page and nothing else on that host: not the function-calling guide, not the SDK source, not the Vertex AI documentation, and nothing here describes them.

There is a written guide: the finishReason dispatch above as a checklist against your own response handling, the tool-schema shapes that give a model the fewest ways to produce an unusable call, and how to make “200 with no usable answer” a branch in your agent loop instead of a silent empty result.

No page on this site has a checkout widget of its own. There is a written guide behind this host and it is on sale at $19 on a storefront that delivers the files automatically and carries a 30-day money-back guarantee: buy it there; the guide can also be paid for with 19 USDC on Base at the payment page, where delivery is by hand as a reply to your email. Every page on this site, including this one, is free to read in full, with no sign-up and nothing gated.

The short version: MALFORMED_FUNCTION_CALL means, in the reference’s own words, that “the function call generated by the model is invalid”. No limit was reached and nothing was counted, so raising max_turns, max_iterations or any other ceiling is wasted effort. It is reported as a field on the response rather than as a request failure, so it looks like an agent that returned nothing. Log the enum value and finishMessage verbatim, keep it distinct from UNEXPECTED_TOOL_CALL and MALFORMED_RESPONSE, and then flatten the tool schema the invalid call was measured against rather than retrying into the same one.

Nearby

finishReason: TOO_MANY_TOOL_CALLS — the neighbouring value on the same field, and the one that is a count: too many consecutive tool calls, stopped server-side.
recursion_limit, max_iter, max_turns: one wall under several names — the caps this page is not, and why raising them is usually the smaller half of a fix even when it is the right half.
Agent stopped due to iteration limit or time limit. — the other ending that hands you something that looks like an answer and is not one.
Agent stopped after 50 turns and produced no output — the same symptom, an empty result, from the opposite cause.
Max number of llm calls limit of `500` exceeded — Google’s ADK ceiling, which does raise and does have a number.

This page counts anonymous readership, and it carries one counter rather than the two that the older pages on this host carry. Each load sends the page path, the address of the page you came from, how long the page was open, whether you scrolled, and any campaign or outreach code in the link you followed; a second “engaged” event is sent once, ten seconds after the page opens — whether or not the tab is in front of you — or as soon as you scroll a quarter of it. The campaign codes from the first link you arrived on are kept in this browser’s local storage, and a later visit that arrives with no codes of its own is counted against them; a link carrying its own codes is used for that visit, and the stored first touch is never replaced. An outreach code is removed from the address bar after it is read. Because this page sends one view event rather than two, a view count taken from it is directly comparable to a load, which is not true of the eighteen pages published before it — those send two, and any rate measured against them reads half its true value. No name is attached to any of this: the only identifier the code can send is an outreach token minted per recipient, and no link carrying one has ever been sent for this page. No cookie; the local storage above does that job. The page also asks this domain for an analytics script at /_vercel/insights/script.js; on 26 September 2026 that address returned HTTP 404 on every host we publish, so no script from another company was served or ran — the page goes on asking, so this stops being true the moment that address starts answering, without a byte of this page changing. Your browser and the network attach things the page does not send: the identification string your browser gives, your IP address and the time of the request. The host that serves this page keeps its own request logs; those are its record and not ours.

“finishReason: MALFORMED_FUNCTION_CALL” — nothing was counted, so nothing can be raised