kanana-2-3b-instruct
kakaocorp/kanana-2-3b-instruct
Official · BF16 · safetensors
Runtime certified: SGLang Transformers vLLM
| Precision / quantization | Format | Bits | Publisher trust | Deployment state |
|---|---|---|---|---|
| BF16 | safetensors | 16 | Official | Runtime certified |
Kanana · General · Kanana 2 SLM
Kanana 2 3B Instruct: Specs, GPU requirements, hosting & deployment
Kanana 2 3B Instruct is a current General release in the Kanana family, certified from its immutable official source revision.
Total parameters3.0B
Active parametersDense model
Context window32,768 tokens
Release statusCurrent · Current
ArchitectureQwen3ForCausalLM
ModalitiesText
LicenseKANANA Open License Agreement
Deployment stateDeployable with certified configuration
Commercial stateCommercial Use Allowed With Conditions
Release date2026-07-24
Architecture, strengths and use cases
Kanana 2 3B Instruct is an official self-hostable Kakao release in the Kanana General track. It uses Qwen3ForCausalLM, carries 3B, supports a verified context window of 32,768 tokens, and exposes only source-certified checkpoints and runtime combinations. Its release, license and deployment facts are tied to the official repository revision rather than inferred from the family name.
Kanana 2 3B Instruct uses Qwen3ForCausalLM with 3B. The stored architecture profile records 32 layers, hidden width 2560, 32 attention heads and a 32,768 token context ceiling where the official configuration publishes those fields.
Verified Qwen3ForCausalLM architecture
32,768 token certified context
Official BF16 checkpoint coverage
Private LLM inference
Agentic workflows
Long-context document processing
Certified checkpoints and quantizations
Only exact, source-certified variants attached to this release are shown.
Recommended hardware
Use only the listed transformers, vllm, sglang runtime configuration with the exact checkpoint precision. The recommendation engine checks weight memory, KV cache, runtime overhead, safety headroom, GPU architecture and topology before ranking rentable and reference hardware.
Estimated weight memory5.76 GB
KV cache1 GB
Runtime overhead5.94 GB
Total with safety margin15.74 GB
Recommended best value
1× NVIDIA A100 SXM 80GB
80 GB total VRAM · nvlink
64.26 GB headroom · Available
USD 915 / monthLowest-cost viable
1× NVIDIA L4 24GB
24 GB total VRAM · pcie
8.26 GB headroom · Available
USD 250 / monthHighest performance
1× NVIDIA H200 SXM 141GB
141 GB total VRAM · nvswitch
125.26 GB headroom · Limited capacity
USD 2,345 / monthLicense and self-hosting considerations
The full 3B resident weight set must fit across the selected GPUs. Context and concurrency add KV-cache memory; MoE active parameters do not replace resident-weight memory.
Kanana 2 3B Instruct is published under KANANA Open License Agreement. The release-scoped official license link and verification date are retained with the catalogue record.
Commercial Use Allowed With Conditions. Commercial use is conditional. Review the linked official terms before deployment.
Alternatives and sibling variants
Kanana 2 1.3B Instruct, Kanana 2 30B-A3B Instruct 2601
Release timeline
kakao/kanana/kanana-2-3b-instruct
Source details
Source revision: d512426765031616ca89fde3a9207f4800d24ccaa319a5dde8d61db0bd4c52b2
Frequently asked questions
How much VRAM does Kanana 2 3B Instruct require?
The exact estimate depends on checkpoint precision, context and concurrency; the page recommendation uses the certified memory engine and compatible runtime configurations.
Which runtimes support Kanana 2 3B Instruct?
Certified upstream runtime evidence is available for transformers, vllm, sglang.
Can Kanana 2 3B Instruct be used commercially?
The stored commercial classification is Commercial Use Allowed With Conditions. Consult the official KANANA Open License Agreement terms for the final obligations.
Official references
Last verified: 2026-08-26