← Catalog
Mi:dm K 3.0 Pro
AvailableKT CorporationProprietary
KT's reasoning model, released Oct 29 2026 on KT's Token Factory platform and listed on OpenRouter on Nov 5. Per KT, a MoE with 120B total and 12B active parameters in a hybrid Mamba / MoE / attention architecture, served as a quantized checkpoint (NVFP4 experts, FP8 Mamba projections and shared expert). Text only, 261,795-token context, $0.20 / $0.80 per 1M input/output tokens. No weights or official benchmarks published so far.
Specifications
- License
- Proprietary · Proprietary
- Weights
- Not released
- Architecture
- hybrid
- Parameters
- 120B · 12B active
- Context window
- 262K tokens
- Max output
- 236K tokens
- Knowledge cutoff
- —
- Price (in / out, $/M)
- $0.2 / $0.8
- Modalities
- Text
Benchmarks
No benchmark scores recorded yet. Spotted some? Submit a correction.
Vendor-reported figures are claims until independently verified. See methodology.