exo Releases

This is a small patch release featuring better sampling defaults, bug fixes for M5 series Macs and RDMA, and support for the new Kimi K2.6 multimodal model.

Apr 23, 2026

Apr 20, 2026

This release introduces multimodality support, significant memory optimizations for long-context use cases, and support for new AI models like Gemma 4 and Qwen3.6.

Apr 17, 2026

This release adds new data tables and query features to the Statistics page, including Valuations Measures and ETFQuery.

Apr 16, 2026

Apr 14, 2026

This release merges missing patches from the previous version, adding currency data to analysis and fixing thread-safety issues.

Apr 13, 2026

This release focuses on Model Context Protocol (MCP) implementation, 3D rendering enhancements, and bug fixes for VRM and Discord integration.

Apr 10, 2026

This release fixes TypeScript type definitions for the sanitizer configuration to accept per-field SanitizerConfig.

Apr 7, 2026

This release introduces continuous batching, Qwen3.5 support, and hardware support for M5 Pro/Max chips, alongside various API and UI improvements.

Mar 27, 2026

This release fixes a bug in the link tool where clicking a formatted link with the Ctrl key now correctly opens the URL in a new window.

Mar 11, 2026

This release addresses a selection change event issue in the inline link tool by utilizing defaultValue to ensure stability.

Mar 4, 2026

The biggest EXO release to date, addressing major stability issues including GPU locks and memory management, while adding extensive new features across models, APIs, and the web dashboard.

Feb 25, 2026

This release fixes a bug where applying bold formatting to linked text would incorrectly remove the link.

Feb 18, 2026

This release improves the unlink behavior in the LinkInlineTool to react more accurately based on the current input state.

Feb 12, 2026

Feb 7, 2026

This release adds support for Kimi-K2.5 model and tensor sharding support for MiniMax M2.1.

Jan 28, 2026

Stability release fixing a regression with RDMA/Tensor Parallelism where models were getting stuck in LOADING state, plus download edge case fixes.

Jan 26, 2026

Stability fixes for RDMA and long running clusters, fixing FAILED -> PREPARING loops, plus QoL features for managing downloads.

Jan 24, 2026

Major release adding GLM-4.7-Flash support, IP-less RDMA discovery, and OpenAI-compatible tool calling via the API, plus bug fixes for auto parallelism.

Jan 23, 2026

Tally introduces enhanced error diagnostics for CSV parsing to help users debug silent failures, alongside Windows test compatibility fixes.

Jan 17, 2026