Canonical: https://botbento.com/blog/gpt-6-astra-tool-calling-responses-api-migration/
Format: Markdown representation of the public HTML page.

[Home](/) / [Blog](/blog/)

FIELD NOTES / 4 MIN READ

# GPT-6 Astra Won't Call Tools on Chat Completions. Now What?

OpenAI's own function-calling guide states that GPT-6 Astra and GPT-6.1 Sol require the Responses API for tool calling, while Chat Completions remains supported for plain text. If your bot's tool-calling code is still on Chat Completions, check which model it's pinned to before you assume an upgrade is a drop-in swap.

By BotBento Editorial · Published 2026-11-05 · Updated 2026-11-05

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](/editorial/).

## In this article

- [Does your bot's tool-calling code need the Responses API?](#direct-answer)
- [What OpenAI's own pages actually say](#what-changed)
- [An illustrative failure path](#worked-example)
- [A decision you can apply today](#migration-checklist)[Read as Markdown ](/text/blog/gpt-6-astra-tool-calling-responses-api-migration/index.md)

## Key takeaways

- OpenAI's function-calling guide states GPT-6 Astra and GPT-6.1 Sol require the Responses API for tool calling; Chat Completions is not deprecated but can't run tools on these models.
- A bot pinned to a dated GPT-5 snapshot keeps working on Chat Completions until OpenAI's December 11, 2026 shutdown of older GPT-5 and o3 snapshots, but silently loses tool access the moment you repoint it at GPT-6 Astra without also changing the endpoint.
- Decide per tool-calling code path, not per app: a flow that only does plain text can stay on Chat Completions; a flow that calls tools on a current-generation model needs the Responses API now.

## Does your bot's tool-calling code need the Responses API?

If any tool-calling code path in your bot points at GPT-6 Astra or GPT-6.1 Sol, yes. OpenAI's function-calling documentation states this directly and without a workaround.

This is a model-level constraint, not a general endpoint deprecation. The same guide treats Chat Completions as still functional for other cases; it just can't carry tool calls on these two models.

The practical trap is that the model page for the previous flagship, GPT-5, tells you to move up a generation without mentioning that the new generation changes which endpoint your tool-calling code has to use. The model recommendation and the endpoint requirement live on two different pages, so an upgrade that looks like swapping a model string can silently break every tool call in that code path.

Sources: [Function calling | OpenAI API](https://developers.openai.com/api/docs/guides/function-calling), [GPT-5 Model | OpenAI API](https://developers.openai.com/api/docs/models/gpt-5).

## What OpenAI's own pages actually say

The function-calling guide is explicit about where the line falls between the two current flagship-tier models and the older Chat Completions examples it still ships. It states that GPT-6 Astra and GPT-6.1 Sol require the Responses API for tool calling, and its Chat Completions sample code is written against an older model with reasoning disabled, not against either of the two current models.

Separately, the GPT-5 model page marks GPT-5 itself as deprecated and tells developers to move to GPT-6 Astra, without restating the tool-calling requirement on that page. Read alone, that page reads like a plain model upgrade.

OpenAI's deprecations page adds the clock: older GPT-5 and o3 model snapshots were given notice on June 11, 2026 for removal from the API on December 11, 2026. That date is about dated snapshots disappearing, not about Chat Completions itself being retired — the guide does not list Chat Completions on that page. Two separate things are true at once: your old model ID has a shutdown date, and your new model has an endpoint requirement. Conflating them is how a planned model bump turns into an unplanned endpoint migration discovered in production.

The pricing page is a smaller but related detail: tokens used for built-in and custom tools are billed at the chosen model's standard per-token input and output rates on either endpoint, so moving from Chat Completions to the Responses API for tool calling does not, by itself, change your token math. Any cost difference comes from caching behavior and conversation handling, not from a different tool-call tariff.

Sources: [Function calling | OpenAI API](https://developers.openai.com/api/docs/guides/function-calling), [GPT-5 Model | OpenAI API](https://developers.openai.com/api/docs/models/gpt-5), [Deprecations | OpenAI API](https://developers.openai.com/api/docs/deprecations), [Pricing | OpenAI API](https://developers.openai.com/api/docs/pricing).

## An illustrative failure path

Illustrative example, not a BotBento test result. Say a small-team bot has a routine that checks a shared calendar and books a meeting room. The tool-calling code was written a year ago against Chat Completions: it sends a messages array, a tools list with a book\_room function, and reads tool\_calls off the response. It has run unattended for months on a pinned GPT-5 snapshot.

A maintainer sees the GPT-5 model page recommending GPT-6 Astra and changes the model string in the client call, nothing else. On the next run, the bot sends the same request shape it always has — messages, tools, tool\_choice — to GPT-6 Astra over Chat Completions. Under the documented requirement, that model cannot take tool calls on that endpoint. The routine either gets an error back or, worse, a response with no tool call in it at all, and the booking step silently never happens.

The fix in this illustrative case is not a prompt tweak. It's an endpoint change: move that one routine's request to /v1/responses, replace the messages array with input items, and read output items instead of a single message.content string. Nothing else about the tool definition has to change. The failure is specific to the endpoint-and-model pairing, which is exactly why testing the old model ID on the new endpoint, and the new model ID on the old endpoint, separately, catches it before a production run does.

Sources: [Function calling | OpenAI API](https://developers.openai.com/api/docs/guides/function-calling).

## A decision you can apply today

Check each tool-calling code path against two questions, in order. First: does this path call GPT-6 Astra or GPT-6.1 Sol? If yes, it needs the Responses API for that path's tool calls — there's no reasoning-effort setting that restores tool calling on Chat Completions for these two models, per OpenAI's guide. Second: if the path is still on an older model, does that model have a shutdown date on OpenAI's deprecations page? GPT-5 and o3 snapshots named in the June 11, 2026 notice stop working December 11, 2026; after that date the first question becomes unavoidable for that path.

If neither applies — the path uses an older model with no scheduled shutdown and calls no tools — there's no forced migration. OpenAI's guidance keeps Chat Completions available for cases outside the tool-calling constraint on these two models; nothing in the sources reviewed here sets a general shutdown date for the endpoint itself.

Treat the model swap and the endpoint swap as two separate changes you test independently, even when you plan to ship them together. Pin the exact model ID you're testing, run the same tool-call fixture against both endpoints where the model supports it, and confirm you're reading the correct output shape (output items on Responses, message.tool\_calls on Chat Completions) before you call the migration done.

Sources: [Function calling | OpenAI API](https://developers.openai.com/api/docs/guides/function-calling), [Deprecations | OpenAI API](https://developers.openai.com/api/docs/deprecations).

## Primary sources

Sources checked 2026-11-05. Standards and product documentation can change; follow the linked version when implementing.

- [Function calling | OpenAI API](https://developers.openai.com/api/docs/guides/function-calling) — OpenAI
- [GPT-5 Model | OpenAI API](https://developers.openai.com/api/docs/models/gpt-5) — OpenAI
- [Deprecations | OpenAI API](https://developers.openai.com/api/docs/deprecations) — OpenAI
- [Pricing | OpenAI API](https://developers.openai.com/api/docs/pricing) — OpenAI

BotBento is in development. [Suggest a correction](/contact/).
