MAX Nightly 26.7.0.dev2026092705 (Mojo 1.2.0.dev2026092705) Released

Nightly
1

:astronaut: A new nightly version has been released! :astronaut:

See the quickstart guide for installation instructions: Quickstart | MAX

MAX changelog updates:

  • DeepSeek-V4 (DeepseekV4ForCausalLM) now runs the released checkpoint
    natively (fp8 linears, fp4 routed experts) and across multiple GPUs, for
    example max serve --devices gpu:0,1. Attention and its indexer split by
    head and the routed experts split across devices (expert parallelism);
    the rest of the model is replicated. The routed experts no longer pad
    every expert group to 128 activation rows, which cuts single-stream
    decode latency. Device graph capture stays off for this architecture.

  • max.driver.DevicePinnedBuffer is removed. Allocate
    Buffer(dtype, shape, device=gpu, usage=Usage.STAGING | Usage.UNTRACKED)
    instead; UNTRACKED keeps the old behavior, where reads never wait on
    device copies. Replace isinstance(b, DevicePinnedBuffer) with b.pinned
    to check for page-locked memory, or Usage.UNTRACKED in b.usage to check
    whether reads skip the hazard wait.

Mojo changelog updates:

  • [mojo-lang] Land exclusivity checking for match.

Raw MAX diff: https://github.com/modular/modular/compare/f23fc09d389bd9826258d5887fa6b2f2c9cbd746...ce67c4bba47a10b8f51b17d07c59336828c3553b)>
Current Mojo changelog: https://github.com/modular/modular/blob/main/Mojo/docs/site/nightly-changelog.md