Pi 1.5.0 adds GPT-6.1 Terra and makes gpt-6.1-sol the default OpenAI model
Pi's v1.5.0 release adds GPT-6.1 Terra for OpenAI, Azure OpenAI Responses and OpenAI Codex, and makes gpt-6.1-sol the default OpenAI model.
Pi, the coding agent from earendil-works, released v1.5.0 on 2 November. For OpenAI users, one change stands out: gpt-6.1-sol replaces gpt-5.5 as the default OpenAI model. The same release adds GPT-6.1 Terra as a selectable model through OpenAI, Azure OpenAI Responses and OpenAI Codex, along with tool search, mid-conversation thinking-level changes and a thinking-off option. The release notes give a 500K context window for Terra on Codex but no window size for the other two routes.
The switch applies even after a ChatGPT sign-in when no model has been selected, so people who never chose a model will get gpt-6.1-sol without doing anything. Pi's changelog ties the change to a single pull request, #11442. The notes do not say whether the new default changes pricing or behavior beyond the model name.
The release also moves the built-in model catalog to models.dev v2. Under that update, realtime and image-only models are no longer offered as chat models, while newer OpenAI and Google chat models are now included. The notes don't list which models were added or removed, so users who depended on a specific entry will need to check the /model list themselves.
The release also includes several other changes. Tool argument validation now runs on @earendil-works/pi-schema, and the release labels this a breaking change. TypeBox features outside JSON Schema, including Type.Refine(), Type.BigInt() and Type.Undefined(), no longer apply to tool arguments. Plain JSON Schema tools, MCP tools among them, now get the same coercion that TypeBox tools already received, such as turning numeric strings into numbers or wrapping a single value in an array. Validation errors that the model sees will now use pi-schema's wording, for example "- path: missing" and "- limit: expected number, got string", in place of TypeBox's "must have required properties path" style.
Extension authors gain a changelog mechanism. Pi reads pi.changelogPath from a package's package.json, or CHANGELOG.md from the package root. After an update, it shows entries newer than the last seen version under "Extension Updates" on the next fresh launch. The collapseChangelog setting condenses that to one line per package. The release also adds a maxBuffer option to pi.exec(), which caps stdout and stderr combined, and a truncated field on the result. It fixes a crash in which a command printing more output than a JavaScript string can hold would bring pi down with a RangeError. Past the default cap of the smaller of 256 MiB and the runtime string limit, the command is terminated.
MCP OAuth sign-in changes how pi registers with authorization servers. By default it now uses pi's Client ID Metadata Document where the server supports it for public clients, and otherwise registers a client dynamically, which the notes compare to what Codex, Claude Code and opencode do. A new setting, oauth.clientRegistration, takes the value "auto" by default, and "dcr" forces registration. Separately, pi now rejects OAuth metadata URLs that are not http: or https:. The notes say that before this fix a malicious MCP server could supply a file: or custom-protocol URL, which pi passed to the platform URL handler.
What changed
Pi's default OpenAI model changes from gpt-5.5 to gpt-6.1-sol, and GPT-6.1 Terra is added as a selectable model.
Type: Breaking change · Applies to: gpt-6.1-sol, gpt-5.5, GPT-6.1 Terra, OpenAI, Azure OpenAI Responses, OpenAI Codex
Why it matters
Teams running Pi with OpenAI accounts will get a different default model after upgrading, which can change cost and output behavior. Codex users gain access to a 500K context window with this model.
What you can do
After upgrading to 1.5.0, check which model your Pi configuration uses, since the OpenAI default has changed.
Original source
Where we saw it
Pi releases
AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it. How it works