BLXBench is built for practical AI model benchmarking, and a lot of that work happens directly in the terminal. Versions 0.6.7 and 0.6.8 focused on making that terminal workflow clearer, faster, and less fragile.
These releases are not about changing what BLXBench measures. They are about reducing friction around how developers start runs, inspect output, submit results, and move through repeated benchmark sessions.
Shell UX: faster navigation while working
Version 0.6.8 improves command recall and transcript navigation inside the interactive shell.
The shell now behaves more like the kind of terminal workflow developers expect:
UpandDownmove through recent slash commandsMeta+UpandMeta+Downmove through command transcriptsPgUpandPgDnwork for output navigation/clearclears the log and command recall while keeping the command itself visible for easy reuse
That matters during repeated benchmark work. You can adjust a run, inspect the result, rerun a related command, and keep moving without fighting the interface.
Clearer TUI feedback
The TUI also received visual and interaction refinements. Error, warning, and info states are easier to distinguish, progress output is easier to scan, and Windows terminal behavior is more predictable around focus and resizing.
These details sound small until you are in the middle of a longer benchmarking session. Clear feedback reduces the chance of missing a failed run, misunderstanding a filter, or losing time to terminal behavior that feels inconsistent.
More predictable filters
The command flow around filters was tightened as well. Commands such as /categories, /levels, and /provider now open interactive pickers directly instead of requiring an extra list step.
Tab completion is more predictable, and selections are constrained to known values where appropriate. That helps prevent half-valid configurations that only become confusing later in the run.
Better report and submission flow
After a run, BLXBench now points more directly to the HTML report, which is usually the file users want to open first. The summary view also supports a fuller in-terminal report mode, so you can inspect results without immediately leaving the TUI.
Manual upload paths remain available even when automatic upload is disabled for a session. This keeps local experimentation and public sharing as separate decisions.
Optional update nudges
BLXBench can also check whether a newer version of @bitslix/blxbench is available when the interactive shell starts. If an update exists, the TUI shows a separate notice instead of mixing that information with sign-in state or run output.
The goal is simple: make staying current visible without making it noisy.
Why these changes matter
Benchmark tooling should stay out of the way. The more predictable the shell and TUI feel, the easier it is to focus on the actual question: how did the models perform, and what do the results mean?
Versions 0.6.7 and 0.6.8 move BLXBench in that direction. They make everyday benchmark sessions smoother without hiding the technical details that developers need to trust the output.
To update BLXBench, use your package manager of choice:
npm update -g @bitslix/blxbench@latest
pnpm update -g @bitslix/blxbench@latest
bun update -g @bitslix/blxbench@latest
You can also review the current product documentation and changelog here: