Mojo 1.0 Release: Key Milestones
Modular released Mojo 1.0 on August 11 through version 26.5, completing a three-year journey since the language's first appearance in 2023. The accompanying MAX 26.5 platform is now the commercial vehicle: Modular itself runs the entire stack of MAX and Modular Cloud on Mojo.
Community contributions since the open-source standard library launched:
- Nearly 200 contributors
- 1,100+ merged pull requests
- 200,000+ lines of code changed
- N-body (500k iterations, FP): CPython 3.14 takes 1242 ms; Mojo takes 16 ms — a 78x speedup
- Spectral-norm: CPython 14046 ms vs Mojo 118 ms — a 119x speedup
- Matrix multiplication: up to 77x speedup
- DNA sequence parsing: Mojo is ~50% faster than Rust
- Large TSV processing and Sudoku solving: Rust still leads
- Asynchronous programming model
- Pattern matching
- Union types
- Source: modular.com/blog/modular-26-5-mojo-1-0-is-here
- Additional reporting: 机器之心 (Jiqizhixin)
- Mojo 1.0 is the first stable release; 1.x will use C++-style backward-compatible evolution.
- Performance vs CPython 3.14: 78x–119x on classical compute benchmarks.
- MLIR compiler enables broader AI-accelerator targeting than LLVM-IR-based Rust.
- Mojo and Rust are complementary, not substitutes, across different CPU workload types.
- Modular targets open-sourcing the compiler and toolchain by 2026.
- 200 contributors, 1,100+ PRs, 200k+ LoC in the open-source standard library.
- N-body benchmark: 1242 ms (CPython 3.14) → 16 ms (Mojo).
- Spectral-norm: 14046 ms → 118 ms.
- DNA parsing: Mojo ~1.5x faster than Rust.
- Large TSV / Sudoku: Rust still leads.
- MAX 26.5 supports GLM-5.2, Nemotron-H; Kimi 2.5 runs on Module V3.
- Agent Skills repo: 7.2K+ downloads on skills.sh.
- Upcoming: async model, pattern matching, union types.
- ModCon scheduled for August 18.
Modular positions 1.0 not as an endpoint but as the beginning of a 1.x phase focused on additive enhancements. Breaking changes will be managed under the discipline used by mature languages such as C++. Developers can now invest in long-term projects without fearing monthly disruptive churn.
Performance Numbers
Mojo's core promise is Python syntax with compiled-native performance. For the first time, that promise has production-grade numbers:
These gains come from three layers: SIMD vectorization, loop unrolling, and an MLIR-based compiler.
Why MLIR Matters
The MLIR layer is especially important for AI workloads. Mojo can target AI accelerators — not only GPUs but also various NPUs and TPU-derived chips — with cross-level optimizations. Rust, born in 2009, is built on LLVM IR and lacks this MLIR pathway.
Benchmark comparisons on CPU tasks:
The two languages are positioned as complementary rather than competing replacements.
Ecosystem and Roadmap
MAX 26.5 ships with native support for GLM-5.2 and Nemotron-H (both hybrid Mamba-2 models), with Kimi 2.5 running on Module V3. The Agent Skills repository from Modular has accumulated 7,200+ downloads (per skills.sh statistics), representing a concrete deployment of MAX/Mojo at the Harness protocol layer.
Roadmap items already on the table:
These are the next building blocks for Mojo's evolution from an AI-specialized language to a general-purpose systems programming language. Modular is also proceeding on a schedule to open-source the Mojo compiler and toolchain by 2026.
ModCon takes place on August 18, where further commercialization and open-source announcements are expected.