Kimi-K2.5
Official · INT4 · safetensors
Runtime certified: Transformers >=4.57.1,<5.0.0 vLLM SGLang KTransformers
| Precision / quantization | Format | Bits | Publisher trust | Deployment state |
|---|---|---|---|---|
| INT4 | safetensors | 4 | Official | Runtime certified |
Kimi · General · Kimi K2.5
Kimi K2.5: Specs, GPU requirements, hosting & deployment
Kimi K2.5 is a previous General release in the Kimi family, certified from its immutable official source revision.
Total parameters1,000.0B
Active parameters32.0B
Context window262,144 tokens
Release statusPrevious · Previous
ArchitectureKimiK25ForConditionalGeneration
ModalitiesText, Image
LicenseModified MIT License
Deployment stateDeployable with certified configuration
Commercial stateCommercial Use Allowed With Conditions
Release date2026-01-01
Architecture, strengths and use cases
Kimi K2.5 is an official self-hostable Moonshot AI release in the Kimi General track. It uses KimiK25ForConditionalGeneration, carries 1000B / 32B MoE, supports a verified context window of 262,144 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.
Kimi K2.5 uses KimiK25ForConditionalGeneration with 1000B / 32B MoE. The stored architecture profile records 61 layers, hidden width 7168, 64 attention heads and a 262,144 token context ceiling where the official configuration publishes those fields.
Verified KimiK25ForConditionalGeneration architecture
262,144 token certified context
Official INT4 checkpoint coverage
Private LLM inference
Agentic workflows
Long-context document processing
Vision-language analysis
Certified checkpoints and quantizations
Only exact, source-certified variants attached to this release are shown.
Recommended hardware
Use only the listed transformers, vllm, sglang, ktransformers 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 memory549.48 GB
KV cache3.34 GB
Runtime overhead156.15 GB
Total with safety margin879.11 GB
Technically compatible reference sizing
No current FusionPointAI inventory passes every runtime, architecture, topology and memory gate for this exact checkpoint. Use the estimator for global compatible hardware.
License and self-hosting considerations
The full 1000B 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.
Kimi K2.5 is published under Modified MIT License. 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
Kimi K3, Kimi K2.6, Kimi-K3
Release timeline
moonshot/kimi/kimi-k2-5 @ 84b793f0055a3ae3b0c7d86ac055b0e8dc190c6e6d548836d00086e96e683be3
Frequently asked questions
How much VRAM does Kimi K2.5 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 Kimi K2.5?
Certified upstream runtime evidence is available for transformers, vllm, sglang, ktransformers.
Can Kimi K2.5 be used commercially?
The stored commercial classification is Commercial Use Allowed With Conditions. Consult the official Modified MIT License terms for the final obligations.
Official references
Last verified: 2026-08-26