Bun AOT Shows 2–3x Speed, Half Memory

// 1h agoBENCHMARK RESULT

Bun AOT Shows 2–3x Speed, Half Memory

Bun creator Jarred Sumner reports that Bun’s experimental ahead-of-time compiler can deliver 2–3x faster performance while using less than half the memory of JIT execution. The approach compiles TypeScript and JavaScript to native code before deployment.

// ANALYSIS

Bun AOT could make JavaScript far more competitive for resource-sensitive CLIs, agents, and server workloads, but it remains experimental and carries meaningful tradeoffs.

–A related Bun benchmark reports 3.7x faster execution, 5.6x lower CPU time, and 2.9x lower memory use versus JIT on one-thread TypeScript workloads. [AOT pull request](https://github.com/oven-sh/bun/pull/45631)
–Claude Code testing showed lower startup time, CPU, and memory, but nearly doubled executable size—an important deployment cost. [Benchmark report](https://aicodingweekly.news/stories/jarred-sumner-bun-aot-claude-code-267mb-67-subagents)
–AOT shifts compilation overhead from every process startup into the build pipeline, making it especially attractive for long-running servers and frequently spawned agent subprocesses.
–Dynamic or weakly typed code may see smaller gains because the compiler relies on inferred runtime shapes and can fall back to slower paths.
–The feature is not yet a stable default, and portability, binary size, and incomplete multithreading support remain open concerns.

// TAGS

bunbun-aotclidevtoolopen-sourcebenchmark

DISCOVERED

1h ago

2026-10-27

PUBLISHED

2h ago

2026-10-27

RELEVANCE

8/ 10

AUTHOR

steipete

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

UPDATE17m ago

PandaCode Connect Outruns T3 Connect, Keeps E2E

PandaCode Connect now reportedly delivers faster steady-state performance than T3 Connect while preserving end-to-end encryption. Reconnect remains slower because its security handshake adds a deliberate latency floor.

LAUNCH21m ago

Query launches secret prompts for coding agents

Query is an open-source macOS utility that lets coding agents request API keys through a floating prompt instead of chat. With `query run`, secrets flow directly into the child process without entering the transcript or agent context.

BENCHMARK46m ago

GPT-6.1 Terra Nearly Matches Sol on Browser Tasks

According to Browser Use’s benchmark update, GPT-6.1 Terra scored 78.5 on browser tasks—nearly matching GPT-6.1 Sol’s 79.7 while costing 35% less. OpenAI positions Terra for fast visual reasoning and precise computer use.