Changelog

Releases

Every release, one page each — newest first. Read the full changelog on a single page, or the raw notes per tag on GitHub.

releases published

Newest 2026-09-30

Latest release

0.15.3 2026-09-30

Supervised Computer Use and more resilient local serving

Read the release notes → or the raw notes per tag on GitHub

Upgrade

$ pip install -U rapid-mlx

$ brew upgrade rapid-mlx

or grab the desktop app.

Earlier releases

Per page

0.15.22026-09-25Easier connections, safer starts, and useful failures
0.15.12026-09-24Failed first starts tell you what went wrong
0.14.32026-09-18Repeat image questions stop re-reading the image, and Desktop can finish small jobs
0.14.22026-09-15Measured decode gains across four model families, and an approval-first Agent mode
0.14.12026-09-11Large and scanned PDFs, and a repetition loop that stops safely
0.14.02026-09-10Measured decode gains, two experimental workspaces, and a much larger catalog
0.13.42026-09-03Qualified Qwen models go faster on their own, and benchmarks stay on your Mac
0.13.32026-09-01GLM-5.3-Flash, and vision chats that stop recomputing
0.13.22026-08-30Flash-Next decodes faster, the app gets 43% smaller
0.13.12026-08-27Qwen3.8-Flash-Next, day-0
0.13.02026-08-26Ornith-1.5, and prefill that tunes itself
0.12.182026-08-21Branch your chat, run more models
0.12.172026-08-19Dictation fixed — update now
0.12.162026-08-19Dictate into any app
0.12.152026-08-18Tool calls stop truncating what your agent writes
0.12.142026-08-15Qwen3.8-27B in our own mixed-precision build
0.12.132026-08-14Qwen3.8-27B, text and vision
0.12.122026-08-13Vision gets its first token faster
0.12.112026-08-12Faster under load, faster alone
0.12.102026-08-11A 131K-context reasoner on an 8 GB Mac
0.12.92026-08-10Muse Glimmer, natively
0.12.82026-08-10The desktop app makes pictures
0.12.72026-08-07One version number
Desktop app 0.12.62026-08-07Rapid can look things up
0.12.52026-08-07Tool calling tells the truth
Desktop app 0.12.52026-08-05An ordinary maths question could take the app down
0.12.42026-08-04Caches that were quietly not caching
0.12.32026-08-04A picker that answers “will this run on my Mac?”
0.12.12026-08-03A desktop app, and a Gemma 4 page that was quoting numbers nobody measured
0.11.92026-08-02Requests that hang, and requests that lie
0.11.82026-08-02Embeddings stop truncating silently at 512 tokens
0.11.72026-08-02Test-only
0.11.62026-08-02DeepSeek V4 gets its own drafter, and a guard against runaway output
0.11.52026-08-01DeepSeek V4 Flash 0731, and a canary for bad output
0.11.42026-07-31Controls for the new multimodal lanes
0.11.32026-07-30Audio grows up, and video ships
0.11.12026-07-28Correctness & first-run polish across the 0.11 line
0.11.02026-07-24Quantized live KV cache — int4/int8 on the continuous-batching cache
0.10.182026-07-24Bounded prefill memory on long prompts
0.10.172026-07-22Base-wheel VLM serve + install-hint fix
0.10.162026-07-22Grammar-constrained Gemma-4 native tool calls
0.10.152026-07-21Grammar-constrained tool calling, out of the box
0.10.142026-07-21Default-on grammar-constrained tool calling
0.10.122026-07-17Response cache + trim-free prefix reuse
0.10.102026-07-15Ternary Bonsai 27B + Qwen3-Coder-Next 80B
0.10.92026-07-12share serve-flag passthrough + MTP K=3 default
0.10.82026-07-11HY3 native MTP (opt-in self-speculative decoding)
0.10.72026-07-10Long-run OOM fix, Hunyuan 3 + Liquid
0.10.52026-07-08Fix batch
0.10.32026-07-07Integrations + parser fixes
0.10.22026-07-07Interpretability
0.10.12026-07-05—
0.10.02026-07-04—
0.9.142026-07-04—
0.9.132026-07-04—
0.9.122026-07-02—
0.9.112026-07-02—
0.9.102026-07-01—
0.9.92026-06-30Multi-user & long-context release
0.8.192026-06-25—
0.8.182026-06-24—
0.8.162026-06-22—
0.8.152026-06-21—
0.8.142026-06-20—
0.8.132026-06-19Audio launch

No releases yet. The raw notes per tag on GitHub stay reachable.