Model comparison

canary-qwen-2.5b vs cohere-transcribe-03-2026

At Q4_K_M, cohere-transcribe-03-2026 is the smaller download — 1,558,162,944 bytes against 1,737,575,808.

From the file· summed bytes, KV per layer

Side by side

canary-qwen-2.5bcohere-transcribe-03-2026
Parameters2.6B2.1B
Architecturecanary_qwencohere_asr
Layers
Native context
Mixture of expertsnono
Quantizations published911
Smallest quantization1.62 GiB1.41 GiB
Q4_K_M1.62 GiB1.45 GiB
Licencecc-by-4.0apache-2.0

KV cache by context

the term that decides long-context viability
Contextcanary-qwen-2.5bcohere-transcribe-03-2026Ratio
4,096
8,192
16,384
32,768
65,536
131,072