Cumulus LabsBack to Leaderboard
LLMs
Qwen 3.5 35B A3B
+28.6%~$QW35M
Hybrid Mamba-SSM and Transformer architecture with 256 experts, 3B active. 128K context with near-zero KV cache memory. FP8 quantized for maximum throughput.
Deploy Now
#13 overall · 2 deploys
1.5s
Cold Start
avg on Ion
215ms
Avg Inference
per request
2
Active Replicas
right now
2
Total Deployments
all time
7-Day Trend
+28.6% this weekCold Start vs Competitors
Lower is better●~$QW35M1.5s
~$QW3-81.2s
~$QW301.7s
~$QW321.8s
Cold start comparison vs similar models. Lower is better.
Cost Estimate
Cost per 1K Inferences
$0.90
Est. Daily (1K req/day)
$0.90
Est. Monthly (30K req)
$27.00
No subscriptions. Buy credits, pay per inference. Scale to zero when idle.
Quick Deploy
cumulus-sdkimport cumulus from "cumulus-sdk" // Deploy Qwen 3.5 35B A3B on Ion const client = await cumulus.deploy("qwen3-5-35b-a3b") // Run inference const result = await client.run({ prompt: "Your prompt here", // model-specific params... })