Release

One exact engine computes every constant, the same at every optimization level, and constants read each other in any order, join strings, hold tables and grow past 64 bits. Characters index arrays and fill sets, a wide integer is written in place, diagnostics go to standard error, the Mica libraries in a C program share one runtime floor and leave their host as it made them, the debugger finds every name at its own place, the editor shows a project file's mistakes where they stand, and 235 defects are closed, each beside a test case, with a program generator checking the compiler every night.

Mica's grammar is published in EBNF with a syntax diagram for every rule, a style guide states how a source is shaped, named, laid out and commented, and mica --format lays a source out in that style. Tasks, generators and processes become dependable from end to end: a program survives the loss of its process server, a deadline reaches a typed file's wait, a generator whose state fits a record pulls at the price of a call, and a technical report explains all three.

A function written over tracked values records its body when the tape window calls it, a model's blocks go to such functions as records, and a library's functions record the same way. The recorded vocabulary gains the logistic, the elementwise product and matrix forms of Exp, Ln and Sqrt; a GPU program carries only the kernels it uses; and fp_contract fuses the same products on both processors at every tier.

A model's whole training step runs on an NVIDIA card in one Mica program: attention over its heads side by side on the tensor cores, the layer normalization, the losses, the update and the clip, each a verb of the language. float16 and bfloat16 stand beside float32 and float64 everywhere a float goes, the numerics are the program's to state, a GPU program carries its kernels precompiled, and the start chooses the card.

Text, arrays and records cross every task, generator, stream and worker process by value; a worker process receives the sockets, pipes and files its task names; a program states how many carriers and workers it runs. C libraries meet Mica as they are written, sqlite is rebuilt on them, a library is linked only under a version, and the editor shows what a stopped program holds.

Mica's mathematics now stands where its systems programming always stood, on the libraries the industry already trusts: the C library on the processor, NVIDIA's compiler, mathematics and cuBLAS on the GPU. Numbers mean what C says they mean at every width, float16 and bfloat16 included. And the release carries everything since 7.5: sum types, tasks in other processes, a runtime that answers a protocol, heapless routines, one call stack across Mica and C, and the editor extension 5.0.

Performance measured on machines anyone can rent, memory brought down, a compiler hardened by an adversarial sweep, by instruments, and by its own gates, and the first technical report - 7.5 is the release where every claim is a measurement, and nothing in it is a new sentence of the language.

The numeric foundation and the training library ship together: tensors whose shapes arrive with the data, any rank, a linear-algebra library on the failure channel, explicit expansion instead of silent broadcasting, a complete training toolkit with visible optimizer state — and the whole vocabulary running resident on an NVIDIA GPU, printing loss columns digit-identical to the host run.

Remarks arrive: hints from the compiler's own knowledge about your code, spoken only when asked and never gating a build — six of them at first light, each explaining a place where Mica is different by design. Beside them, the callback kind meets the C world's callback-shaped APIs, libraries learn to export generators and borrowed handles across their contracts, and VS Code extension 4.1.0 puts all of it in the editor: remarks in the Problems panel with a status-bar counter, and the snippet family returns.

Training a network becomes ordinary Mica: values that record onto the language's own autograd tape, one Backward walk answering every gradient, GPU residency checked at compile time, and formulas written with their own glyphs. Shipped together with VS Code extension 4.0 — Mica Notebooks whose outputs are verified facts, one-click run and debug, and a stopped program that narrates its own task tree. A complete GPT trains in pure Mica, and a twelve-chapter course teaches the road to it.