✨ Gemma 4 is here! ✨

Read about the release in the blog post

3.19.0 (2026-06-30)

Features

• Gemma 4 support (#591) (5fe6e27) (documentation: Gemma 4)
• riscv64 prebuilt binaries (#615) (e8336a4)
• automatically enable flash attention when optimal
• improve inference performance when a grammar is active
• more precise resource usage estimation
• resource usage capping (documentation: Resource Capping)
• automatically enable or disable mmap depending on the environment
• support Q1_0 quant
• improve stability on unified memory systems
• disable residency sets on macOS by default for better OS responsiveness
• default progressLogs to "stderr" to avoid polluting stdout with logs
• optimized prebuilt binaries for arm architectures

Bug Fixes

• MXFP4_MOE quant name
• Vulkan backend successful load detection even when no devices are available
• CLI: avoid redownloading existing models that consist of multiple parts from a URI
• optimize checkpoints management when using grammar
• improve stability when loading huge models
• reranking result range for Qwen 3 reranker
• adapt to breaking llama.cpp changes

Shipped with llama.cpp release b9842

To use the latest llama.cpp release available, run npx -n node-llama-cpp source download --release latest. (learn more)

3.18.1 (2026-03-17)

Features

• customize postinstall behavior (#582) (57bea3d) (documentation: Customizing postinstall Behavior)
• experimental support for context KV cache type configurations (#582) (57bea3d) (documentation: LlamaContextOptions["experimentalKvCacheKeyType"])
• support NVFP4 quants (#582) (57bea3d)

Shipped with llama.cpp release b8390

To use the latest llama.cpp release available, run npx -n node-llama-cpp source download --release latest. (learn more)

3.18.0 (2026-03-15)

Features

• automatic checkpoints for models that need it (#573) (c641959)
• QwenChatWrapper: Qwen 3.5 support (#573) (c641959)
• inspect gpu command: detect and report missing prebuilt binary modules and custom npm registry (#573) (c641959)

Bug Fixes

• resolveModelFile: deduplicate concurrent downloads (#570) (cc105b9)
• correct Vulkan URL casing in documentation links (#568) (5a44506)
• Qwen 3.5 memory estimation (#573) (c641959)
• grammar use with HarmonyChatWrapper (#573) (c641959)
• add mistral think segment detection (#573) (c641959)
• compress excessively long segments from the current response on context shift instead of throwing an error (#573) (c641959)
• default thinking budget to 75% of the context size to prevent low-quality responses (#573) (c641959)

Shipped with llama.cpp release b8352

To use the latest llama.cpp release available, run npx -n node-llama-cpp source download --release latest. (learn more)

3.17.1 (2026-02-28)

Bug Fixes

• Electron template (#566) (8931402)

Shipped with llama.cpp release b8179

To use the latest llama.cpp release available, run npx -n node-llama-cpp source download --release latest. (learn more)

3.17.0 (2026-02-27)

Features

• getLlama: build: "autoAttempt" (#564) (dda5ade) (documentation: LlamaOptions ["build"])
• remove octokit dependency (#564) (dda5ade)

Bug Fixes

• CLI: disable Direct I/O by default (#564) (dda5ade)
• Bun segmentation fault on process exit with undisposed Llama instance (#564) (dda5ade)
• detect glibc inside Nix (#564) (dda5ade)

Shipped with llama.cpp release b8169

To use the latest llama.cpp release available, run npx -n node-llama-cpp source download --release latest. (learn more)

3.16.2 (2026-02-21)

Bug Fixes

• macOS 14 prebuilt binaries (#559) (6faa5ae)

Shipped with llama.cpp release b8121

To use the latest llama.cpp release available, run npx -n node-llama-cpp source download --release latest. (learn more)

3.16.1 (2026-02-20)

Bug Fixes

• export missing types (#557) (498711c)

Shipped with llama.cpp release b8117

To use the latest llama.cpp release available, run npx -n node-llama-cpp source download --release latest. (learn more)

3.16.0 (2026-02-19)

Features

• Exclude Top Choices (XTC) (#553) (57e8c22) (documentation: LLamaChatPromptOptions["xtc"])
• DRY (Don't Repeat Yourself) repeat penalty (#553) (57e8c22) (documentation: LLamaChatPromptOptions["dryRepeatPenalty"])
• Tiny Aya support (#553) (57e8c22)

Bug Fixes

• adjust the default VRAM padding config to reserve enough memory for compute buffers (#553) (57e8c22)
• support function call syntax with optional whitespace prefix (#553) (57e8c22)
• change the default value of useDirectIo to false (#553) (57e8c22)
• Vulkan device dedupe (#553) (57e8c22)

Shipped with llama.cpp release b8095

To use the latest llama.cpp release available, run npx -n node-llama-cpp source download --release latest. (learn more)

3.15.1 (2026-01-26)

Bug Fixes

• adapt to llama.cpp changes (#547) (4baa480)
• duplicate backend library files (#541) (f5123bf)

Shipped with llama.cpp release b7836

To use the latest llama.cpp release available, run npx -n node-llama-cpp source download --release latest. (learn more)

3.15.0 (2026-01-10)

Features

• LlamaCompletion: stopOnAbortSignal (#538) (734693d) (documentation: LlamaCompletionGenerationOptions["stopOnAbortSignal"])
  LlamaModel: useDirectIo (#538) (734693d) (documentation: LlamaModelOptions["useDirectIo"])

Bug Fixes

• support new CUDA 13.1 archs (#538) (734693d)
• build the prebuilt binaries with CUDA 13.1 instead of 13.0 (#538) (734693d)

Shipped with llama.cpp release b7698

To use the latest llama.cpp release available, run npx -n node-llama-cpp source download --release latest. (learn more)