$ diff fugu-max kimi-k3
Fugu Max vs Kimi K3
Values come from Fugu Max and Kimi K3, 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.
Sakana AIFugu Max
- context
- 1M
- weights
- —
- $/M in
- $2
- $/M out
- $6
- context
- 1M
- weights
- open
- $/M in
- $3
- $/M out
- $15
What actually differs
- Context window
- Identical — both accept 1M tokens, so context length is not a reason to pick either.
- Input price
- Fugu Max at $2/M against $3/M — 1.5× cheaper to feed. Standard rates: one of these labs also quotes a lower cached-input tier, which applies only when a prefix is reused — see the full spec below.
- Output price
- Fugu Max at $6/M against $15/M — 2.5× cheaper to generate. Output dominates the bill on most agentic workloads, where the model writes far more than it reads.
- Recency
- Fugu Max shipped 57 days after Kimi K3 (2026-09-11 vs 2026-07-16).
Full spec
| Attribute | Fugu Max | Kimi K3 |
|---|---|---|
| Developer | Sakana AI | Moonshot AI |
| Released | 2026-09-11 | 2026-07-16 |
| Context window | 1,000,000 | 1M tokens |
| Pricing | $2/M input · $6/M output | $0.30/M cache-hit input · $3.00/M cache-miss input · $15.00/M output |
| License | not recorded | Modified MIT (commercial use permitted; weights publicly downloadable) |
| Availability | Sakana API (OpenAI-compatible), OpenRouter, NanoGPT, Kilo Gateway, LLM Gateway | Kimi Code, Kimi app, Kimi API; weights on Hugging Face (`moonshot-ai/kimi-k3`, released 2026-07-27) |