Top Generation Speed
Peak
38.45
tokens/sec
Sovereign Cluster Record
Lowest TTFT
Latency
0.850
seconds
Time To First Token
Total Submissions
Verified
6
runs
Ingested via Feed & Web
Leading Model
Architecture
qwen2.5-coder-32b-instruct
Most Frequent Benchmark
Automate Your Own Telemetry
Run our standardised one-line Python script on your host to benchmark local LLM speed and auto-submit to this feed:
curl -s https://ai.enarah.net.au/benchmark_agent.py | python3 - --alias "YourAlias" --code ENARAH-PUBLIC --submit
Live Benchmark Leaderboard
Ordered by generation speed (TPS) descending
| Rank / Alias | Model | Hardware / Backend | TTFT | Gen Speed | Tokens | Submitted |
|---|---|---|---|---|---|---|
|
1
Sentinel-Spark
|
qwen3.6:35b-a3b |
NVIDIA GB10 130GB
Ollama / CUDA
|
0.850s | 38.45 tps | 324 comp | Aug 25, 2026 13:28 |
|
2
Sentinel-Ultra
|
google/gemma-4-31b-qat |
Apple M3 Ultra 103GB
LM Studio / MLX
|
0.940s | 31.20 tps | 316 comp | Aug 25, 2026 13:28 |
|
3
Mabel-M2
|
qwen2.5-coder-32b-instruct |
Unknown Host
N/A
|
0.950s | 29.80 tps | 200 comp | Aug 25, 2026 14:20 |
|
4
Test-Agent-M2
|
qwen2.5-coder-32b-instruct |
Apple Mac mini M2 Pro 32GB
Local Test Client
|
1.150s | 29.71 tps | 312 comp | Aug 25, 2026 13:28 |
|
5
Sentinel-Hera
|
qwen2.5-coder-32b-instruct |
Apple M3 Ultra 103GB
LM Studio / MLX
|
1.209s | 27.68 tps | 312 comp | Aug 25, 2026 13:28 |
|
6
Test-Client-CLI
|
google/gemma-4-31b-qat |
Test MacBook Pro
Ollama / CPU
|
1.450s | 22.40 tps | 206 comp | Aug 25, 2026 13:46 |