MAX Nightly 26.7.0.dev2026102205 (Mojo 1.2.0.dev2026102205) Released

Nightly
1

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

See the quickstart guide for installation instructions: Quickstart | MAX

MAX changelog updates:

  • InternVL and Gemma 3 multimodal models now return log probabilities, so
    logprobs in max serve and output_logprobs in generate() work for
    them.

  • /v1/completions now returns a 400 error for echo with logprobs on a
    server started without --enable-echo, when the model computes log
    probabilities. The request used to crash the model worker. Other models
    still return no log probabilities, and echo without logprobs has no
    effect.

  • MAX Serve’s span attributes now sit under the entity they describe:
    max.request_id is now max.request.id, max.batch_id is max.batch.id,
    and the other request and batch attributes move under max.request.* and
    max.batch.*. kernel_agg:<name> spans carry the kernel name in
    max.kernel.op instead of max.kernel.name; per-kernel spans keep
    max.kernel.name. Update trace queries that filter on the old names.

  • max.nn.Transformer, DistributedTransformer, logits_postprocess() and
    distributed_logits_postprocess() return a
    max.nn.graph_groups.DecoderOutputs of named groups instead of a tuple.
    Read the logits as
    outputs.logits["logits"] rather than outputs[1], and pass the outputs to
    Graph.output(outputs) rather than unpacking them. The graph’s outputs keep
    their order, and Graph.output_tree returns them by name.

  • ops.matmul() takes an out_dtype keyword argument that sets the result
    dtype. The inputs keep their own dtypes and are no longer cast to a common
    one, so bfloat16 inputs with out_dtype=DType.float32 keep the float32
    accumulator. Without out_dtype, ops.matmul(), the @ operator and
    max.experimental.functional’s matmul() require both inputs to share a
    dtype, which the result keeps, so max.nn.Linear raises when its input and
    weight dtypes differ.
    max.experimental.nn.Linear casts its input to the weight’s dtype, so a
    float32 input to a bfloat16 Linear now gives bfloat16 instead of float32.

  • Removed the blocking_sync, disable_timing and interprocess parameters
    from the Mojo DeviceContext.create_event(), along with EventFlags. Events
    never record timing; use DeviceContext.execution_time() to time kernels.

Mojo changelog updates:

  • [stdlib] Move AMD-only intrinsics to max.gpu.intrinsics
  • [mojo-compiler][stdlib] Deprecate @explicit_destroy on structs
  • [mojo-lang][stdlib] Remove the legacy kernel capture ABI (#105477)
  • [stdlib] Add an owning FileDescriptor type
  • [mojo-lang][stdlib] Remove legacy closures (#105361)
  • [mojo-lang] Fix import leaks from precompiled packages
  • [stdlib] stabilize some more list methods.
  • [stdlib] Fix the SIMD.slice fast-path offset check

Raw MAX diff: https://github.com/modular/modular/compare/a1ccd09545970f73bd33527ef027d445b6ad0980...c783ebd98c26aa012cee07ecee96de212f19f22e
Current Mojo changelog: https://github.com/modular/modular/blob/main/Mojo/docs/site/nightly-changelog.md