Model family
Two models, one Turkish recipe
1.0 Max is the flagship that competes on the board. 1.0 Mini is the same data and the same tokenizer, distilled until it runs on hardware you already own.
Yiğido 1.0 (Max)
FlagshipFull mixture-of-experts checkpoint. The model behind every number on the benchmark page.
- 32B total / 4B active params
- 128K context window
- Agent Arena rank 12 · +4.80% net
- ~180 tok/s on 8× H100
Yiğido 1.0 Mini
EdgeDistilled sibling for laptops, phones and single-GPU servers — including the machine running this site.
- 9B total / 1.4B active params
- 32K context window
- Agent Arena rank 18 · +3.42% net
- ~420 tok/s on one 24 GB card
Side by side
Pick by hardware, not by hype
Model
1.0 Max
1.0 Mini
Total / active params
32B / 4B
9B / 1.4B
Context
128K tokens
32K tokens
Net improvement
+4.80%
+3.42%
Confirmed success
8.60%
6.05%
Steerability
2.10%
1.15%
Latency
~180 tok/s (8× H100)
~420 tok/s (1× 24 GB)
VRAM (int4)
~20 GB
~6 GB
License
Open weights
Open weights
Benchmark figures are from the Agent Arena Overall snapshot of Sep 15, 2026. Latency measured with vLLM, 1,024-token prompt, 256-token completion.
Talk to the first Turkish AI
Open weights, a drop-in API and a chat window that runs on modest hardware. Start with a question in Turkish.