Anomaly, the company behind open-source coding agent OpenCode, shipped Hex 3.0 stable: hold a shortcut, speak, and the text is pasted at your cursor. Recognition runs on-device, local model memory drops from about 900 MB to about 160 MB, and it adds vocabulary hints, a hex CLI and optional OpenCode rewriting.
Key Takeaways
- ✓Hex v3.0.0 was released on GitHub at 00:38 UTC on Oct 31; it is MIT-licensed, written in Rust, and the repo has about 1,557 stars (release notes)
- ✓Vendor figures: local model memory drops from about 900 MB to about 160 MB; since 2.3.0 it streams, so a 2-minute dictation finishes about 0.1 s after release on an RTX 5070 Ti
- ✓Runs on macOS 15+ (Apple silicon), Windows x64/ARM64 and Linux x64/ARM64, with new .deb and .rpm packages on Linux
- ✓3.0 does not auto-update from 2.3.1 and starts fresh, so settings, History and models must be set up again; Voice Action and spoken commands are removed
- ✓The @opencode launch post drew about 2,400 likes, 1,600 bookmarks and 100k+ views (post)

Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background and pain point
Coding agents keep getting more capable, so the bottleneck is shifting to how clearly people can describe what they want. Typing long specs and review notes is slow, and most dictation tools send audio to the cloud, which is awkward when you are naming internal APIs or customers. On Oct 31, Anomaly, the company behind open-source coding agent OpenCode, announced the stable release of its Hex dictation app from the @opencode account, arguing that voice is the best way to talk to any agent.
How it works
Hex is native Rust: hold a shortcut, speak, release, and the text is pasted wherever your cursor is. Transcription runs entirely on the device with a downloaded local model such as Parakeet or Whisper, and ordinary dictation needs neither OpenCode nor a cloud service. Version 3.0 adds a vocabulary list to fix names and terms, a hex CLI for History and meeting notes with an MCP entry point, and optional rewriting through a running OpenCode instance.
Benchmarks and measured results
No standardized accuracy benchmark has been published; the per-model accuracy shown in the picker is the vendor's own. The official figures that can be checked: local models now use about 160 MB of memory versus about 900 MB before, and since 2.3.0 a 2-minute dictation finishes about 0.1 s after release, versus 0.9 s on an RTX 5070 Ti and 2.8 s on a MacBook Air M4 before.
Developer takeaways
On macOS, install with brew install --cask anomalyco/tap/hex; Windows and Linux have install scripts plus .deb and .rpm packages at anoma.ly/hex. Note that 3.0 does not update itself from 2.3.1 and starts fresh, so settings, History and models must be set up again, and macOS needs Microphone and Accessibility permissions granted again. It is worth a try for developers who dictate long requests to Claude Code, Codex or OpenCode.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.