$ diff kimi-k2-8-preview granite-4-2
Kimi K2.8 Preview vs Granite 4.2
Values come from Kimi K2.8 Preview and Granite 4.2, where each is cited to its source. This page states no benchmark result and ranks neither model — it puts two published specifications next to each other. Where a lab has not published a figure, the row says so rather than guessing.
Moonshot AIKimi K2.8 Preview
- context
- 1M
- weights
- closed
- $/M in
- —
- $/M out
- —
- context
- 512K
- weights
- open
- $/M in
- —
- $/M out
- —
What actually differs
- Context window
- Kimi K2.8 Preview takes 1M against 512K — modestly more room in a single request.
- Weights
- Granite 4.2 publishes weights (Apache 2.0); Kimi K2.8 Preview is API-only. That decides self-hosting, air-gapped deployment and fine-tuning before any capability question does.
- Recency
- Kimi K2.8 Preview shipped 17 days after Granite 4.2 (2026-09-11 vs 2026-08-25).
Full spec
| Attribute | Kimi K2.8 Preview | Granite 4.2 |
|---|---|---|
| Developer | Moonshot AI | IBM |
| Released | 2026-09-11 | 2026-08-25 |
| Context window | 1,000,000 | 512K (30B); unknown (3B, 8B) |
| Pricing | not recorded | not applicable — open weights, no hosted rate published |
| License | proprietary | Apache 2.0 |
| Availability | Kimi API platform, Kimi Code, Kimi Work | Hugging Face, Ollama, GitHub |