Pi v0.79.9 introduces native thinking controls

// 1h agoPRODUCT UPDATE

Pi v0.79.9 introduces native thinking controls

Pi v0.79.9 updates the open-source terminal coding harness to map agent thinking levels into chat template kwargs for OpenAI-compatible providers, allowing models like DeepSeek to use provider-native thinking controls. The release also improves session switching and deep branching performance, resolves GLM-5.2 routing, filters unavailable GitHub Copilot models, and fixes bugs with WSL bash expansion, streaming markdown fences, and fuzzy code-editing matches.

// ANALYSIS

This update represents a major step forward in making Pi robust for local reasoning models and long-running developer sessions.

–Mapping thinking levels directly to vLLM and Hugging Face chat-template parameters enables native support for deep reasoning models like DeepSeek.
–Speeding up deep session branches from quadratic to linear time ensures the agent remains responsive during long coding sessions.
–Preserving untouched code lines in fuzzy matches reduces the risk of incorrect file rewrites, a common issue in agentic file-editing pipelines.

// TAGS

agentcoding-assistantsdevtoolopen-sourcedeepseekreleases

DISCOVERED

1h ago

2026-06-21

PUBLISHED

1h ago

2026-06-21

RELEVANCE

6/ 10

AUTHOR

PiChangelog

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

BENCHMARK16m ago

User praises GLM-5.2 as frontier open-weight model

A Twitter post highlights the impressive performance of the open-weight model GLM-5.2, noting it as a true frontier model. The author then asks when a Gemini model will appear in the top 3, presumably referring to a popular LLM leaderboard.

NEWS1h ago

IIT Bombay, BharatGen join Project Tapestry

India's sovereign AI initiative BharatGen, led by IIT Bombay, has joined the AI Alliance's Project Tapestry as a founding contributor and country anchor. The collaboration will see BharatGen co-lead workstreams in distributed training to shape globally federated, open-source frontier AI models.

NEWS1h ago

AI stigma on Steam drops game reviews 53%

A data analyst has observed that the presence of an "AI stigma" on the Steam platform can significantly impact a game's reception, reducing the overall number of reviews it receives by approximately 53%. Furthermore, the reviews that these AI-associated games do manage to get tend to be more negative compared to standard games.

Pi v0.79.9 introduces native thinking controls