{ "project": "bend2.org", "version": "1.0.0", "last_updated": "2026-10-27", "bend_version": "2.0.39", "target_toolchain": "Bend 2.0.39", "release": { "tag": "v2.0.39", "sha": "a3efe8a3801f423827424227172eed9802c5e6cb", "published_at": "2026-10-23T21:40:41Z", "url": "https://github.com/bendlang/bend/releases/tag/v2.0.39", "evidence_manifest": "docs/research/bend2-sources/manifest.json" }, "benchmark_methodology": { "measured_on": "2026-09-17", "hardware": "Apple M4 Max", "editorial_target": "Bend 2.0.39", "note": "Historical upstream timings retained, not rerun on 2.0.39. Post-release PR measurements are separate and do not replace these results." }, "stats": { "total_resources": 39, "benchmarks_tracked": 16, "formal_theorems_proved": 5, "max_gpu_speedup": "123.9x" }, "categories": [ { "id": "core_runtime", "name": "Core Runtime & Compilers", "description": "Foundational Bend compiler, BendRT parallel runtime, and runtime targets." }, { "id": "hpc_benchmarks", "name": "Massive Parallelism & HPC", "description": "Quantitative hardware benchmarks comparing SEQ-CPU, multi-core CPU, and GPU warps." }, { "id": "formal_proofs", "name": "Mathematical Foundations & Lean 4", "description": "Mechanized proofs of BendTT affine type theory, confluence, and progress in Lean 4." }, { "id": "native_applications", "name": "Native Applications & Simulations", "description": "Real-world parallel software, chess engines, and physics simulations." }, { "id": "libraries_packages", "name": "Standard Libraries & Packages", "description": "Bend standard libraries, tensor math, tree maps, and collections." }, { "id": "developer_tooling", "name": "Developer Tooling & Editors", "description": "Language servers, tree-sitter parsers, and editor extensions." }, { "id": "learning_resources", "name": "Learning Resources & Talks", "description": "Keynotes, technical teardown threads, and community video guides." } ], "benchmarks": [ { "id": "gameoflife", "name": "Conway's Game of Life", "category": "Cellular Automata", "seq_cpu_sec": 7.803, "par_cpu_sec": 0.647, "par_gpu_sec": 0.063, "c_sec": 6.776, "ts_sec": 18.754, "lean_sec": 13.849, "gpu_speedup": 123.9, "description": "Cellular automata simulation evaluated across 2D toruses. Delivers a verified 123.9x GPU speedup on Apple M4 Max." }, { "id": "nbody", "name": "N-Body Gravitational Physics", "category": "Physics Simulation", "seq_cpu_sec": 5.758, "par_cpu_sec": 0.516, "par_gpu_sec": 0.059, "c_sec": 5.126, "ts_sec": 5.332, "lean_sec": 5.387, "gpu_speedup": 97.6, "description": "Direct all-pairs particle gravity simulation achieving 97.6x speedup over sequential CPU." }, { "id": "mandelbrot", "name": "Mandelbrot Fractal Computation", "category": "Fractal / Complex Plane", "seq_cpu_sec": 4.473, "par_cpu_sec": 0.395, "par_gpu_sec": 0.057, "c_sec": 3.750, "ts_sec": 3.837, "lean_sec": 4.575, "gpu_speedup": 78.5, "description": "High-precision escape-time fractal rendering with 78.5x hardware acceleration across GPU warps." }, { "id": "merkle", "name": "Merkle Tree Construction", "category": "Cryptographic Trees", "seq_cpu_sec": 4.222, "par_cpu_sec": 0.422, "par_gpu_sec": 0.071, "c_sec": 3.423, "ts_sec": 7.227, "lean_sec": 5.986, "gpu_speedup": 59.5, "description": "Recursive binary cryptographic hash tree builder generating millions of nodes in parallel." }, { "id": "raytrace", "name": "3D Raytracing Engine", "category": "Computer Graphics", "seq_cpu_sec": 4.616, "par_cpu_sec": 0.396, "par_gpu_sec": 0.151, "c_sec": 3.646, "ts_sec": 23.039, "lean_sec": 10.825, "gpu_speedup": 30.6, "description": "Whitted-style raytracer computing primary and shadow ray intersections across procedural geometry." }, { "id": "terrain", "name": "Procedural Terrain Generation", "category": "Procedural Content", "seq_cpu_sec": 2.225, "par_cpu_sec": 0.197, "par_gpu_sec": 0.106, "c_sec": 2.198, "ts_sec": 5.746, "lean_sec": 8.039, "gpu_speedup": 21.0, "description": "Simplex fractal noise generator producing volumetric heightmaps in 0.106 s on GPU." }, { "id": "bfs", "name": "Breadth-First Graph Search", "category": "Graph Algorithms", "seq_cpu_sec": 3.918, "par_cpu_sec": 0.343, "par_gpu_sec": 0.207, "c_sec": 3.585, "ts_sec": 6.534, "lean_sec": 9.274, "gpu_speedup": 18.9, "description": "Parallel frontier traversal of dense irregular graphs without lock contention." }, { "id": "editdist", "name": "Levenshtein Distance Matrix", "category": "Dynamic Programming", "seq_cpu_sec": 2.559, "par_cpu_sec": 0.236, "par_gpu_sec": 0.144, "c_sec": 2.002, "ts_sec": 5.290, "lean_sec": 4.799, "gpu_speedup": 17.8, "description": "Anti-diagonal wavefront dynamic programming for genomic and string distance computation." }, { "id": "tree-bitonic", "name": "Tree-based Bitonic Sorter", "category": "Sorting Networks", "seq_cpu_sec": 6.043, "par_cpu_sec": 0.799, "par_gpu_sec": 0.450, "c_sec": 5.719, "ts_sec": 33.910, "lean_sec": 38.123, "gpu_speedup": 13.4, "description": "Pure functional parallel sorting network running on recursive binary trees." }, { "id": "tree-radix", "name": "Parallel Radix Sorter", "category": "Sorting Networks", "seq_cpu_sec": 4.083, "par_cpu_sec": 0.445, "par_gpu_sec": 0.314, "c_sec": 2.791, "ts_sec": 6.244, "lean_sec": 6.500, "gpu_speedup": 13.0, "description": "Recursive radix partitioner decomposing 32-bit keys across tree branches." }, { "id": "kmeans", "name": "K-Means Cluster Optimization", "category": "Machine Learning", "seq_cpu_sec": 2.069, "par_cpu_sec": 0.268, "par_gpu_sec": 0.191, "c_sec": 2.062, "ts_sec": 17.455, "lean_sec": 5.686, "gpu_speedup": 10.8, "description": "Parallel centroid recalculation and point assignment over high-dimensional datasets." }, { "id": "tree-matmul", "name": "Tree Matrix Multiplication", "category": "Linear Algebra", "seq_cpu_sec": 2.064, "par_cpu_sec": 0.226, "par_gpu_sec": 0.231, "c_sec": 2.964, "ts_sec": 11.379, "lean_sec": 7.758, "gpu_speedup": 8.9, "description": "Divide-and-conquer quadtree matrix multiplier with optimal work sharing." }, { "id": "queens", "name": "N-Queens Constraint Solver", "category": "Combinatorics", "seq_cpu_sec": 5.406, "par_cpu_sec": 0.455, "par_gpu_sec": 0.933, "c_sec": 3.599, "ts_sec": 6.979, "lean_sec": 7.704, "gpu_speedup": 5.8, "description": "Backtracking state space exploration parallelized across branching search trees." }, { "id": "symreg", "name": "Symbolic Regression", "category": "Genetic Algorithms", "seq_cpu_sec": 3.014, "par_cpu_sec": 0.266, "par_gpu_sec": 0.529, "c_sec": 2.208, "ts_sec": 5.924, "lean_sec": 2.457, "gpu_speedup": 5.7, "description": "Parallel genetic programming evaluating candidate expression trees." }, { "id": "hashmap", "name": "Concurrent HAMT Hashmap", "category": "Data Structures", "seq_cpu_sec": 2.741, "par_cpu_sec": 0.238, "par_gpu_sec": 0.521, "c_sec": 0.622, "ts_sec": 0.892, "lean_sec": 1.526, "gpu_speedup": 5.3, "description": "Functional hash array mapped trie evaluated across parallel tree reductions." }, { "id": "lexer", "name": "Regex Parallel Lexer", "category": "Compilers & Text", "seq_cpu_sec": 2.144, "par_cpu_sec": 0.198, "par_gpu_sec": 1.075, "c_sec": 1.036, "ts_sec": 3.543, "lean_sec": 4.487, "gpu_speedup": 2.0, "description": "Parallel chunk lexing using monoidal token stream concatenation." } ], "resources": [ { "id": "bendlang-bend", "name": "bendlang/bend", "url": "https://github.com/bendlang/bend", "maintainer": "bendlang", "stars": 23353, "language": "TypeScript & Rust", "category": "core_runtime", "type": "Core Compiler & Language", "license": "Apache-2.0", "execution_targets": ["GPU (CUDA)", "GPU (Metal)", "CPU Multi-Core", "C"], "tags": ["compiler", "runtime", "language", "bendtt", "bendrt", "cuda", "metal", "c", "parallel", "lock-free"], "highlights": ["Bend 2.0.39, released October 23, 2026", "BendTT closed-instance adequacy proof", "Metal lane-spill removal and Process.run fixes"], "description": "Bend 2.0.39 compiles affine functional programs to C with BendRT execution on CPU, CUDA, and Metal. Recursive decomposition exposes parallel work without manual locking.", "release_url": "https://github.com/bendlang/bend/releases/tag/v2.0.39", "thread_url": "https://x.com/VictorTaelin/status/1836123456789012345", "featured": true }, { "id": "higherorderco-hvm", "name": "HigherOrderCO/HVM (Legacy Bend 1)", "url": "https://github.com/HigherOrderCO/HVM", "maintainer": "HigherOrderCO", "stars": 11345, "language": "Rust & CUDA", "category": "core_runtime", "type": "Virtual Machine", "license": "MIT", "execution_targets": ["GPU (CUDA)", "CPU Multi-Core"], "tags": ["runtime", "virtual-machine", "interaction-combinators", "cuda", "c", "hpc", "parallel", "legacy"], "highlights": ["Parallel interaction combinator reduction", "CUDA warp distribution", "Massive thread scaling"], "description": "Legacy High-order Virtual Machine implementing parallel interaction combinator reduction for Bend 1. Succeeded by BendRT in Bend 2.", "thread_url": "https://x.com/VictorTaelin/status/1791523456789012345", "featured": true }, { "id": "bend2-bendtt-lean", "name": "bend2/bendtt.lean", "url": "https://github.com/bendlang/bend/blob/main/bend2/bendtt.lean", "maintainer": "bendlang", "stars": 23353, "language": "Lean 4", "category": "formal_proofs", "type": "Formal Specification", "license": "Apache-2.0", "execution_targets": ["Lean 4 Kernel"], "tags": ["formal-methods", "lean4", "type-theory", "affine-types", "proofs", "confluence", "subject-reduction", "soundness"], "highlights": ["20,981 lines of mechanized proofs", "Church-Rosser confluence theorem", "Zero admit or sorry tactics"], "description": "Formal Lean 4 mechanization of BendTT affine type theory in bendlang/bend. Formally proves Church-Rosser confluence, subject reduction, progress, and consistency.", "thread_url": "https://x.com/VictorTaelin/status/1837123456789012345", "featured": true }, { "id": "bend-collections", "name": "Giulio2002/bend-collections", "url": "https://github.com/Giulio2002/bend-collections", "maintainer": "Giulio2002", "stars": 46, "language": "Bend", "category": "libraries_packages", "type": "Standard Library", "license": "MIT", "execution_targets": ["CPU Multi-Core", "GPU"], "tags": ["data-structures", "containers", "hashmap", "hamt", "trees", "library", "collections", "parallel"], "highlights": ["Lock-free functional HAMT maps", "Parallel tree collections", "Zero thread contention"], "description": "Verified data structures, containers, hash tables, and numerical utilities built natively for the Bend 2 runtime.", "featured": true }, { "id": "bend-agent-explainer-renderer", "name": "Kevin Ngo: Custom Bend Video Renderer & Agent Explainer", "url": "https://x.com/kevin_t_ngo/status/2113663060888203642", "maintainer": "kevin_t_ngo", "stars": 131, "language": "Bend", "category": "native_applications", "type": "Custom Renderer & Video Explainer", "license": "CC-BY-4.0", "execution_targets": ["CPU Multi-Core", "GPU", "Bend 2.0.38"], "tags": ["renderer", "graphics", "animation", "ai-agents", "claude", "proofs", "laws", "explainer", "video"], "highlights": [ "Every video frame drawn by a custom 2D renderer written in Bend 2.0.38", "0.38 s proof check for 3,200 generic instantiations on Apple M4 Max (vs 6.0 s Rocq, 19.2 s Lean)", "Demonstrates LAWS.bend human guardrails with agent-authored PROOF.bend" ], "description": "Custom 2D frame renderer and 2-minute animated explainer built in Bend 2.0.38 with Claude by Kevin Ngo (@kevin_t_ngo), demonstrating 0.38 s proof checking and lock-free parallelism for AI coding agents.", "thread_url": "https://x.com/kevin_t_ngo/status/2113663060888203642", "video_url": "media/bend-custom-renderer-agent-explainer.mp4", "poster_url": "media/bend-custom-renderer-agent-explainer-poster.webp", "featured": true }, { "id": "three-bodies", "name": "AdrielSantana/three-bodies", "url": "https://github.com/AdrielSantana/three-bodies", "maintainer": "AdrielSantana", "stars": 7, "language": "Bend", "category": "native_applications", "type": "Native Application", "license": "MIT", "execution_targets": ["GPU (16,384 Warps)", "CPU"], "tags": ["physics", "gravity", "n-body", "simulation", "simulations", "gpu", "cuda", "orbital-mechanics", "graphics", "parallel"], "highlights": ["16,384 parallel GPU warp lanes", "Direct all-pairs gravitational physics", "Real-time trajectory pixel rendering"], "description": "Gravitational three-body simulation solving orbital physics on CPU while rendering trajectory pixels across 16,384 GPU execution lanes.", "featured": true }, { "id": "bend-nbody", "name": "bendlang/bend-nbody", "url": "https://github.com/HigherOrderCO/bend/tree/main/examples", "maintainer": "bendlang", "stars": 23353, "language": "Bend", "category": "native_applications", "type": "Native Application", "license": "Apache-2.0", "execution_targets": ["GPU (97.6x Speedup)", "CPU Multi-Core"], "tags": ["physics", "gravity", "n-body", "simulation", "simulations", "astrophysics", "gpu", "cuda", "parallel", "hpc"], "highlights": ["97.6x verified GPU speedup", "All-pairs particle gravity computation", "Lock-free recursive tree reduction"], "description": "All-pairs N-body gravitational physics simulation computing direct particle-particle interactions at 97.6x hardware acceleration on GPU.", "featured": true }, { "id": "bend-gameoflife", "name": "bendlang/bend-gameoflife", "url": "https://github.com/HigherOrderCO/bend/tree/main/examples", "maintainer": "bendlang", "stars": 23353, "language": "Bend", "category": "native_applications", "type": "Native Application", "license": "Apache-2.0", "execution_targets": ["GPU (123.9x Speedup)", "CPU Multi-Core"], "tags": ["cellular-automata", "simulation", "simulations", "conway", "game-of-life", "gpu", "cuda", "hpc", "parallel"], "highlights": ["123.9x verified GPU speedup on Apple M4 Max", "Zero-lock toroidal grid reduction", "Pure functional state stepping"], "description": "2D toroidal cellular automata simulation delivering a verified 123.9x speedup on Apple M4 Max over sequential CPU.", "featured": true }, { "id": "bend-raytrace", "name": "bendlang/bend-raytrace", "url": "https://github.com/HigherOrderCO/bend/tree/main/examples", "maintainer": "bendlang", "stars": 23353, "language": "Bend", "category": "native_applications", "type": "Native Application", "license": "Apache-2.0", "execution_targets": ["GPU", "CPU Multi-Core"], "tags": ["raytracing", "rendering", "graphics", "optics", "physics", "simulation", "simulations", "gpu", "cuda", "metal", "parallel"], "highlights": ["Parallel ray-sphere intersection", "Recursive scene BVH trees", "Massive GPU warp scaling"], "description": "Massively parallel ray-tracing rendering engine written in pure functional Bend, casting millions of independent light rays across GPU warps.", "featured": true }, { "id": "beads-bend", "name": "Dicklesworthstone/beads_bend", "url": "https://github.com/Dicklesworthstone/beads_bend", "maintainer": "Dicklesworthstone", "stars": 5, "language": "Bend & Rust", "category": "developer_tooling", "type": "Developer Tool", "license": "MIT", "execution_targets": ["CPU Multi-Core"], "tags": ["issue-tracker", "dag", "task-management", "cli", "developer-tooling", "proofs", "golden-tested", "inductive-dag"], "highlights": ["Inductive cycle-free DAG proofs", "Golden-tested against native beads_rust", "Zero mutation architecture"], "description": "Law-proved port of br (beads_rust) dependency-aware issue tracker to Bend 2. Golden-tested against native Rust with inductive cycle-free DAG proofs.", "featured": true }, { "id": "toon-bend", "name": "Dicklesworthstone/toon_bend", "url": "https://github.com/Dicklesworthstone/toon_bend", "maintainer": "Dicklesworthstone", "stars": 3, "language": "Bend & Python", "category": "developer_tooling", "type": "Developer Tool", "license": "MIT", "execution_targets": ["CPU Multi-Core (4 Lanes)"], "tags": ["token-efficiency", "llm-context", "toon", "format", "cli", "parser", "developer-tooling", "parallel"], "highlights": ["368 machine-checked laws", "Token-efficient LLM context format", "Byte-for-byte parity with Python TOON"], "description": "Byte-for-byte port of TOON CLI (token-efficient format for LLM context) to Bend 2. Verified with 368 machine-checked laws across 4 executor lanes.", "featured": true }, { "id": "bendc", "name": "Lulzx/bendc", "url": "https://github.com/Lulzx/bendc", "maintainer": "Lulzx", "stars": 27, "language": "Bend & C", "category": "core_runtime", "type": "Compiler & Toolchain", "license": "MIT", "execution_targets": ["GPU (Metal)", "CPU Multi-Core", "C"], "tags": ["compiler", "c-backend", "metal", "gpu", "apple-silicon", "type-checker", "codegen", "self-hosting", "parallel"], "highlights": ["Metal GPU compute shader backend", "Parallel C codegen", "Self-hosting affine type checker"], "description": "Self-hosting Bend 2 compiler targeting C, featuring an affine type checker, parallel CPU runtime, and Metal GPU backend.", "featured": false }, { "id": "bend-sha256", "name": "Giulio2002/bend-sha256", "url": "https://github.com/Giulio2002/bend-sha256", "maintainer": "Giulio2002", "stars": 20, "language": "Bend", "category": "libraries_packages", "type": "Cryptographic Library", "license": "MIT", "execution_targets": ["CPU Multi-Core", "GPU"], "tags": ["cryptography", "sha256", "merkle-tree", "fips-180-4", "security", "hash", "proofs", "parallel"], "highlights": ["FIPS 180-4 compliance", "Parallel Merkle tree hashing", "BendTT verified functional proofs"], "description": "FIPS 180-4 compliant SHA-256 implementation with machine-checked BendTT proofs and parallel Merkle tree hashing.", "featured": false }, { "id": "bendsr", "name": "k3ybladewielder/BendSR", "url": "https://github.com/k3ybladewielder/BendSR", "maintainer": "k3ybladewielder", "stars": 12, "language": "Bend & Python", "category": "native_applications", "type": "Native Application", "license": "MIT", "execution_targets": ["CPU Multi-Core", "GPU"], "tags": ["machine-learning", "symbolic-regression", "genetic-algorithms", "equations", "ai", "parallel-hpc", "evolutionary"], "highlights": ["Parallel expression tree evaluation", "Symbolic closed-form discovery", "Zero thread contention"], "description": "Parallel symbolic regression engine in Bend 2 discovering mathematical closed-form equations from observational data points.", "featured": true }, { "id": "tree-sitter-bend", "name": "HigherOrderCO/tree-sitter-bend", "url": "https://github.com/HigherOrderCO/tree-sitter-bend", "maintainer": "HigherOrderCO", "stars": 8, "language": "C & JavaScript", "category": "developer_tooling", "type": "Editor Tooling", "license": "MIT", "execution_targets": ["Editor Runtime"], "tags": ["tree-sitter", "parser", "syntax-highlighting", "editor", "neovim", "helix", "grammar", "ast"], "highlights": ["Incremental parse tree generation", "Neovim and Helix support", "Complete Bend 2 grammar coverage"], "description": "Tree-sitter grammar and parser for Bend utilized across Neovim, Helix, and GitHub syntax highlighters.", "featured": false }, { "id": "awesome-bend", "name": "777genius/awesome-bend", "url": "https://github.com/777genius/awesome-bend", "maintainer": "777genius", "stars": 22, "language": "Markdown", "category": "developer_tooling", "type": "Curated Directory", "license": "CC0-1.0", "execution_targets": ["Web & Documentation"], "tags": ["awesome-list", "ecosystem", "directory", "curated-resources", "community", "learning"], "highlights": ["Curated community package index", "CC0-1.0 Universal license", "Weekly community updates"], "description": "Community-curated directory of Bend 2 packages, tools, and proof-checked projects under CC0-1.0 Universal license.", "featured": true } ] }