v3.8.0

Released September 30, 2026 · View on GitHub →

TL;DR — GPT-6.1 Sol arrives as the new OpenAI default: near-Astra quality at $2/$10, with cache reads at half GPT-6 Sol's price. A setting on GPT-6 Sol keeps it.

Added

at $2/$10, the same as GPT-6 Sol, with cached input at half the price ($0.10), a 1.05M context and 128K output. Its thinking levels run from low to max; it has no "none", and Codeep never sends it one. Through OpenRouter it is openai/gpt-6.1-sol.

on "Send terminal answers to Telegram" in /settings, and the "Codeep finished" message carries the agent's reply, as it already did for a task sent from the phone. Off by default: an answer can include a file the agent read or a command line, and the chat is stored on Telegram's servers.

Changed

to OpenAI. A setting already on GPT-6 Sol keeps it, including one that got GPT-6 Sol as the default in 3.6 or 3.7, so pick GPT-6.1 Sol with /model if you want it. GPT-6 Sol stays in the picker as the previous Sol.

Astra, it calls tools only over OpenAI's Responses API, which Codeep uses at OpenAI's own endpoint. With OPENAI_BASE_URL set (Azure, LiteLLM and the like) or CODEEP_OPENAI_WIRE_API=chat, its agent turns use Codeep's text tool format, and Codeep says so once. A proxy that serves /v1/responses gets native tool calls back with CODEEP_OPENAI_WIRE_API=responses. GPT-6 Sol still calls tools natively there, with reasoning off, and the notice names it.

Downloads

Install with npm install -g codeep@3.8.0.