Unsloth · Qwen 3.5
Qwen 3.5 35B A3B GGUF
Verified by Badgr · ServeQwen 3.5 35B A3B GGUF is a 35B chat and text-generation model available through supported Badgr execution routes.
Last validated or meaningfully updated: 2026-10-10
✓ Verified by Badgr · llama.cpp · Serve
Last verified 2026-10-10
badgr serve --runtime llama.cpp --hf-repo unsloth/Qwen3.5-35B-A3B-GGUF --hf-file Qwen3.5-35B-A3B-UD-Q3_K_XL.gguf --gpu RTX_4090 --max-cost 2 --startup-timeout 40 --env LLAMA_ARG_CTX_SIZE=8192 --env LLAMA_ARG_N_GPU_LAYERS=99 --env LLAMA_ARG_MMPROJ_URL=https://huggingface.co/unsloth/Qwen3.5-35B-A3B-GGUF/resolve/main/mmproj-F16.gguf- GPU
- NVIDIA GeForce RTX 4090Reported by the provider, not confirmed inside the container
- Time to verified
- 330s
- Deployment
- dep-39927bb860
- ✓ llama-server started with the vision projector loaded (/props reports vision: true)
- ✓ A 200-token chat completion returned
- ✓ final_state = VERIFIED
- ✓ Teardown completed
Q3_K_XL with the F16 projector served on CUDA at about 180 tokens/s decode in a single request. ggml-org/llama.cpp#30230 reports about 0.1 tokens/s for this quant plus projector on a Vulkan iGPU; that setup was not tested here. The elapsed time is estimated from the spend at verification and the hourly rate.
GitHub issue this run was for: ggml-org/llama.cpp#30230
Raw evidence
final_state=VERIFIED failure_class=None providers_tried=1 spend_usd=0.04 P30230 completion_tokens=200 decode_tok_per_s=179.7 prefill_tok_per_s=138.0
Run from Badgr's local development environment. Documents that this deployment path works end-to-end for this model.
Availability
Not yet Badgr AI API
Not currently validated
✓ Dedicated endpoint
Available through Badgr
✓ Custom GPU deployment
Available through Badgr
Deployment profile
- Architecture
- qwen35moe (GGUF)
- Licence
- Apache-2.0
- Minimum VRAM
- ~74GB
- Context
- 256K tokens
Recommended: A100 80GB. VRAM is an estimate and increases with context, cache, concurrency, and runtime overhead.
Model identity
- Base model
- Qwen/Qwen3.5-35B-A3B
- Updated
- 7 months ago
- HuggingFace revision
- bc014a17be
Popularity and trust
Downloads (last month)
913.4K
Likes
866
Spaces using this model
13
Files and formats
Weight formats
GGUF
Repository files
32
Chat template
Not detected
GPU deployment scenarios
Estimated from parameter count and quantisation. Will switch to “Verified by Badgr” once a real run backs a tier.
Testing
A100 80GB
Short context, low concurrency
Small production
H100 80GB
8K context, 1-4 concurrent requests
Higher throughput
H200 141GB
32K context, continuous batching
Large production
A100 80GB
Higher concurrency
Estimated — multimodal inputs, long context, and concurrency all increase real VRAM use beyond this estimate.
Also runs well on H100 80GB, H200 141GB. Available in United States, Europe, Asia Pacific.
Serving compatibility
| Runtime | Status | Notes |
|---|---|---|
| vLLM | Estimated | Used for Badgr dedicated endpoints |
| Transformers | Unknown | Basic fallback |
| llama.cpp | Verified by Badgr | Confirmed by a real badgr serve run (deployment dep-39927bb860) |
| SGLang | Unknown | Not evaluated |
| Ollama | Unknown | Not evaluated |
| Diffusers | Unknown | Not evaluated |
| PyTorch | Unknown | Not evaluated |
| TensorRT-LLM | Unknown | Conversion may be required |
| TGI | Unknown | Not evaluated |
Ready-to-run Badgr configurations
Quick test
badgr serve unsloth/Qwen3.5-35B-A3B-GGUF --max-cost 2Validated dedicated endpoint command
badgr serve unsloth/Qwen3.5-35B-A3B-GGUF --gpu A100 80GB --max-cost 10Advanced configuration (defaults to automatic)
- Runtime
- Quantisation
- GPU and GPU count
- Context length
- Maximum concurrency
- Region
- Maximum hourly spend
- Persistent or capped runtime
Estimated pricing
Provider estimate: cold start
3–8 min
Provider estimate: endpoint cost
$1.40 – $1.40/hr
Idle cost
$0 when stopped
Per-token throughput cost is not shown here: Badgr has not benchmarked this model yet, and this page does not display figures it cannot back with real data.
Recommended Badgr routes
A100 80GB · United States
Estimatedstarting from $1.40/hr
Badgr estimated starting price
Best for
LoRA Training, vLLM Endpoint, Qwen 2.5 32B Instruct
Startup: 3–8 min
Reliability: High
Last checked: Live marketplace price not available yet
A100 80GB · Europe
Estimatedstarting from $1.40/hr
Badgr estimated starting price
Best for
LoRA Training, vLLM Endpoint, Qwen 2.5 32B Instruct
Startup: 3–8 min
Reliability: Good
Last checked: Live marketplace price not available yet
A100 80GB · Europe
Estimatedstarting from $1.40/hr
Badgr estimated starting price
Best for
LoRA Training, vLLM Endpoint, Qwen 2.5 32B Instruct
Startup: 3–8 min
Reliability: High
Last checked: Live marketplace price not available yet
A100 80GB · Europe
Estimatedstarting from $1.40/hr
Badgr estimated starting price
Best for
LoRA Training, vLLM Endpoint, Qwen 2.5 32B Instruct
Startup: 3–8 min
Reliability: High
Last checked: Live marketplace price not available yet
A100 80GB · Europe
Estimatedstarting from $1.40/hr
Badgr estimated starting price
Best for
LoRA Training, vLLM Endpoint, Qwen 2.5 32B Instruct
Startup: 3–8 min
Reliability: Good
Last checked: Live marketplace price not available yet
A100 80GB · Europe
Estimatedstarting from $1.40/hr
Badgr estimated starting price
Best for
LoRA Training, vLLM Endpoint, Qwen 2.5 32B Instruct
Startup: 3–8 min
Reliability: High
Last checked: Live marketplace price not available yet
Known limitations
- Memory use increases with context length and concurrency.