Experimental & Cross-Host Languages — Tier 3 Index

Experimental & Cross-Host Languages — Tier 3 Index

  • Type: Family index
  • Family: Modern experimental systems languages + transpiled / hosted languages
  • Languages catalogued: ~28
  • Last updated: 2026-05-07

Family overview

This index covers two adjacent populations that share a common trait: they target a host runtime, a host language, or a future they haven’t quite reached yet.

The first population — modern experimental systems languages — comprises the post-Rust generation of pre-1.0 contenders. Each picks a different design axis to push: Carbon prioritises C++ migration, Vale and Austral push memory models (regions, capabilities), Mojo bets on Python compatibility plus MLIR, Hare bets on minimalism, Bend on implicit parallelism. Most are pre-1.0 in 2026 and most will probably stay there — the systems-programming niche is brutally competitive and Rust’s gravitational pull is strong. The interesting question is which one or two will break through, and on what dimension.

The second population — transpiled / hosted languages — compiles to a host runtime (JS, JVM, .NET, Wasm, EVM, MoveVM) rather than to native machine code. This is a strategic choice rather than a technical limitation: targeting JavaScript means instant deployment to every browser; targeting the JVM means access to the Java ecosystem; targeting Wasm means portability without a sandbox; targeting the EVM means writing Ethereum smart contracts. The languages in this group span an enormous range — CoffeeScript was a syntactic sugar, ReScript is a real ML descendant, Solidity is the dominant smart-contract language, AssemblyScript is TypeScript-flavoured AOT-to-Wasm, Dafny is verification-flavoured C#-on-.NET.

The two populations meet in blockchain languages (Move, Solidity, Vyper, Cairo, Sway) — these are simultaneously experimental (the runtimes are <10 years old) and host-targeted (they compile to a specific VM bytecode) — and in verification languages (Dafny, F*) which target a host runtime but exist primarily as research vehicles.

The state of these populations in 2026 is bifurcated. The systems-experimental group has had no breakthrough since Rust, but Mojo is growing rapidly in the AI/ML niche and Carbon has Google’s backing. The transpiled group is dominated by TypeScript (which has its own deep note — it became the JS ecosystem) and the blockchain languages (Solidity remains dominant on Ethereum; Move is the second platform thanks to Aptos and Sui). Most of the smaller transpiled-to-JS languages (CoffeeScript, LiveScript, Imba) have been steamrolled by TS+ESNext.

In our deep library

  • v — V (covered deeply)
  • odin — Odin (covered deeply)
  • roc — Roc (covered deeply)
  • gleam — Gleam (covered deeply)
  • pony — Pony (covered deeply)
  • typescript — TypeScript (the dominant transpiled language; covered deeply)
  • rust — Rust (the gravitational pull behind every modern systems-lang attempt)
  • zig — Zig (sibling to many in this list)
  • ada — Ada and SPARK (the verification ancestor)
  • ml-and-fp — covers Elm, PureScript, ReasonML, F*

Tier 3 — the family

Modern experimental systems languages (pre-1.0)

LanguageFirst releaseStatus 2026NicheWhy it mattersSource URL
Carbon2022Pre-1.0 (Google)C++ successor with bidirectional interopGoogle-backed; explicit C++-migration storyhttps://github.com/carbon-language/carbon-lang
Vale2020Pre-1.0Region-based memory; “single ownership without GC, without borrow checker”Novel generational references; explores memory models past Rusthttps://vale.dev/
Hare2022Pre-1.0Minimalist systems langDrew Devault; “C without the worst footguns”; small spechttps://harelang.org/
Jakt2022ExperimentalSerenityOS systems lang (Andreas Kling)Used to gradually replace C++ in SerenityOShttps://github.com/SerenityOS/jakt
Austral2022ResearchCapability-based + linear typesLinear types as primary discipline; very small corehttps://austral-lang.org/
Mojo2023Active growthPython-superset for AI/ML; MLIR-basedModular Inc.; targets Python ecosystem; SIMD/GPU-first; fastest-growing in this grouphttps://www.modular.com/mojo
Wing (Winglang)2022ActiveCloud-first languageCompiles to TypeScript+Terraform; multi-cloud abstractionhttps://www.winglang.io/
Inko2015 (1.0 in 2024)ActiveActor-based concurrentPony-influenced; gradual typing → static; reached 1.0https://inko-lang.org/
Beef2019Active (game-dev)Game-dev systems langBuilt by BeefyTech for game development; C#-flavoured but nativehttps://www.beeflang.org/
Bend2024Pre-1.0Implicit parallelism via HVMHigh-level lang that runs on Higher-Order Virtual Machine; auto-parallel on GPU/CPUhttps://higherorderco.com/
Ante2018ResearchAlgebraic effects + refinement types”Low-level lang with high-level features”; effect systemhttps://antelang.org/
Skew2014DormantSmall/embeddedCompiles to JS, C++, C#; mostly Evan Wallace’s researchhttps://skew-lang.org/
Onyx2021Active (small)Modern systems; Wasm-friendlyUsed in indie game-dev community; MIT-licensedhttps://onyxlang.io/
Cone2018ResearchReference capabilities for memory safetyJonathan Goodwin; explores Rust-alt memory modelshttps://cone.jondgoodwin.com/
Lobster2010Active (small)Game-dev experimentalWouter van Oortmerssen (also Flatbuffers); built for game scriptinghttp://strlen.com/lobster/

Transpiled / hosted (compiles to another runtime)

LanguageFirst releaseStatus 2026Host targetNiche / Why it mattersSource URL
CoffeeScript2009LegacyJavaScriptPioneered “lang-on-JS”; killed by ES6+/TShttps://coffeescript.org/
LiveScript2011DormantJavaScriptCoffeeScript with FP flavour; some Haskell influencehttps://livescript.net/
Iced CoffeeScript2012DormantJavaScriptAsync additions to CoffeeScript before async/await landed in JShttps://maxtaco.github.io/coffee-script/
ReScript (formerly BuckleScript)2020 (renamed)ActiveJavaScriptOCaml→JS compiler; React-friendly; strong types; used at Facebook originhttps://rescript-lang.org/
Reason2016Mostly absorbed by ReScriptJavaScript / OCamlOCaml-with-JS-syntax; original Facebook React Reason efforthttps://reasonml.github.io/
Imba2015NicheJavaScriptDOM-aware web language; tag-as-syntax; small but activehttps://imba.io/
Stencil2017Active (Ionic)JavaScript / Web ComponentsTS-flavoured compiler for Web Components; used by Ionic Frameworkhttps://stenciljs.com/
Marko2014 (eBay)ActiveJavaScripteBay’s UI lang; SSR-first; competitive perfhttps://markojs.com/
AssemblyScript2017ActiveWebAssemblyTypeScript subset → AOT Wasm; common Wasm authoring choicehttps://www.assemblyscript.org/
Grain2017Active (small)WebAssemblyML-flavoured; Wasm-native; designed for browser+server Wasmhttps://grain-lang.org/
Motoko2020ActiveInternet Computer (DFINITY)Actor-based; built for the IC blockchain; persistent memory modelhttps://internetcomputer.org/docs/current/motoko/
Move2019ActiveMove VM (Aptos, Sui, originally Diem)Resource-oriented type system; #2 smart-contract platform after EVMhttps://move-language.github.io/move/
Solidity2015Active (dominant)Ethereum (EVM)The Ethereum smart-contract language; runs ~$200B+ TVLhttps://soliditylang.org/
Vyper2017ActiveEthereum (EVM)Python-flavoured; security-focused; security-conscious EVM alternativehttps://docs.vyperlang.org/
Cairo2020 (StarkWare)ActiveCairo VM (Starknet)Zero-knowledge-proof-native; STARK-friendlyhttps://www.cairo-lang.org/
Sway2022ActiveFuelVM (Fuel blockchain)Rust-flavoured smart-contract language for Fuel L2https://fuellabs.github.io/sway/
Dafny2009 (Microsoft Research)Active.NET / JavaScript / Go / PythonVerification-first; compiles to multiple hosts; used at AWS, MSRhttps://dafny.org/
F* (FStar)2011 (Microsoft Research / Inria)ActiveOCaml / F# / C / WasmDependently typed verification; used in HACL* (Firefox NSS, Linux kernel crypto)https://www.fstar-lang.org/
Rune2020Active (small)Rust embeddingRust-flavoured embeddable scripting; Lua-style use case for Rust appshttps://rune-rs.github.io/

Notable threads

  • Why “C++ successor” is so crowded. C++ committee glacial-pace evolution + a $1T installed base + Rust’s not-quite-drop-in interop creates a market gap that everyone wants to fill. Carbon (Google), Cppfront (Herb Sutter), Circle (Sean Baxter), and many others are all variations on “give me modern type safety with C++ ABI compatibility.” None has clearly won yet; Carbon is the bet with the most institutional weight.
  • Mojo is the breakout candidate of the experimental group. Modular’s Mojo achieves something rare: it is a Python superset (any Python program is a Mojo program), but adds value types, SIMD primitives, MLIR-based codegen, and GPU targeting. The AI/ML niche makes this a serious commercial play — every PyTorch user is a potential Mojo user. It’s the only post-Rust experimental systems language with a credible path to mass adoption as of 2026.
  • The blockchain language explosion. Solidity (2015) dominated unchallenged for years. Then Move (2019, originally Diem/Facebook, now Aptos and Sui) introduced a resource type system — values that cannot be implicitly copied or dropped — which prevents an entire class of token-duplication bugs. Cairo (2020) brought zero-knowledge-proof-native programming. Vyper offers a more conservative EVM alternative. The lesson: smart-contract languages have to bake in formal-methods ideas because bugs cost millions and are unfixable post-deployment.
  • Verification languages are quietly winning where it matters. F*‘s HACL* library is in Firefox’s NSS, the Linux kernel, and Microsoft Edge. Dafny is used at AWS for distributed-systems verification (the AWS S3 strong consistency proof was Dafny). These languages are tiny in user-count but enormous in deployed lines verified. Verification went from “research curiosity” to “what serious crypto and distributed-systems teams use” between 2015 and 2025.
  • The transpiled-to-JS graveyard. CoffeeScript at peak (~2013) was on every Rails project. By 2017 ES6+ had absorbed every interesting feature (arrow functions, destructuring, let/const, classes, template literals) and TypeScript had absorbed the rest plus added types. Today CoffeeScript, LiveScript, Iced, and most of their cohort are vestigial. The lesson: betting against a host language that is itself improving is a losing trade. ReScript survived because OCaml-strength types are a thing JS will never get.
  • Why most experimental systems languages stay pre-1.0. Rust raised the bar on what “production systems language” means: a working borrow checker, a working async story, a working package manager, a working ecosystem, multi-platform support, ABI stability commitments, an RFC process. Reaching that bar takes 8-10 years and a paid team. Most of the languages in the experimental list have a paid team of 0-3 people. Mojo and Carbon have well-resourced teams; most others don’t.
  • The “pre-1.0 forever” risk. Several languages in this list (Vale, Austral, Cone, Bend) are research-flavoured and may never converge on 1.0 — that’s not a failure mode, that’s the point. They demonstrate ideas that get absorbed into the next generation of mainstream languages. Vale’s region references, Austral’s linear-as-default, Cone’s reference capabilities — these will likely show up in some mainstream language in the 2030s, just not necessarily under those names.

Citations