Qwen3.8 27B
Alibaba · China · Apache 2.0
Commercial use: Yes — free for commercial use
Apache 2.0, no conditions. Note that Alibaba did NOT extend Apache 2.0 to the whole family: Qwen3.8-Flash-Next and the 2.4T-A95B flagship both ship under a custom "other" licence. The 27B is the permissive one.
The most capable model that fits comfortably on a single high-end graphics card, released under Apache 2.0.
Why it should matter to you
This is the realistic starting point for a business that wants its own model rather than an API. One 24 GB card, no licence conditions, good enough at reasoning and code to be genuinely useful. If someone has told you private AI needs a rack of hardware, this is the counter-example.
How it connects to our work
Apache 2.0 is also the licence that survives legal review without a conversation. No user thresholds, no attribution clause, no publisher right to restrict you later — unlike Llama and Gemma, which are widely and wrongly described as open source.
From our field notesName the software: Ollama to serve it, vLLM at concurrency, Open WebUI as the screen.
Advantages
- The single most-downloaded current open model on the Hub (6.2M+ across BF16 and FP8 repos)
- Apache 2.0 with no conditions, on a size that fits one 24 GB card
- The best capability-per-gigabyte available on consumer hardware right now
- Enormous fine-tune and quantisation ecosystem — every runtime supports it on day one
Disadvantages
- Dense at 27B, so slower per token than a comparable MoE
- The larger Qwen3.8 models are not Apache 2.0, so scaling up within the family changes your licence
- Thinking mode roughly triples token usage if you leave it on for simple tasks
Reach for it when
The default local model for most businesses. One 24 GB GPU or a 32 GB Mac, no licence conditions, good enough at code and reasoning for real work.
Where it falls down
Very long multi-document reasoning where a frontier MoE pulls ahead, and high-concurrency serving where a sparse model of similar quality would be cheaper per request.
Running it
24 GB GPU at Q4_K_M, or a 32 GB Mac at Q6_K. See the hardware sizing tables for how that maps to specific chips and cards, and the quantisation guide for what you give up at each bit width.
Which cloud instance actually runs this, and what it costs
At Q4_K_M the weights are about 16 GB, so 20 catalogued instances fit with room for context, of which the 8 best value are shown. Cheapest per token is Hetzner GEX45 (~$1.92 per million output tokens at an estimated 42 tok/s). Fastest on a single card is Azure NCads H100 v5 at roughly 209 tok/s for $9.26 per million tokens, which is the speed-versus-cost trade in one line.
| Instance | GPU | VRAM | Bandwidth | Est. tok/s | $/hr | $/M tokens |
|---|---|---|---|---|---|---|
| Hetzner GEX45 | RTX PRO 4000 Blackwell | 24 GB | 672 GB/s | ~42 | $0.290 | $1.92 |
| RunPod A100 PCIe | A100 80GB | 80 GB | 1935 GB/s | ~121 | $1.190 | $2.73 |
| Hetzner GEX131 | RTX PRO 6000 Blackwell Max-Q | 96 GB | 1792 GB/s | ~112 | $1.420 | $3.52 |
| RunPod L40S | L40S | 48 GB | 864 GB/s | ~54 | $0.790 | $4.06 |
| RunPod H100 PCIe | H100 PCIe | 80 GB | 2000 GB/s | ~125 | $1.990 | $4.42 |
| Lambda A100 40GB | A100 40GB | 40 GB | 1555 GB/s | ~97 | $1.990 | $5.69 |
| Lambda H100 PCIe | H100 PCIe | 80 GB | 2000 GB/s | ~125 | $3.290 | $7.31 |
| AWS g5.xlarge | A10G | 24 GB | 600 GB/s | ~38 | $1.006 | $7.45 |
Token generation on a single request is memory-bandwidth-bound, not compute-bound: the GPU must read every active weight from memory for each token it emits. So the ceiling is roughly GPU memory bandwidth divided by the size of the active weights. A 16 GB model on a 300 GB/s L4 tops out near 19 tokens/sec; the same model on a 2039 GB/s A100 tops out near 127. Figures on model pages are that arithmetic, computed from the bandwidth and weight-size columns here. They are a ceiling for one request at batch size 1, before serving overhead, so treat them as an upper bound for comparing instances rather than a benchmark. Batching raises total throughput well above this and does not raise per-request speed.
For identical silicon, the hyperscalers charge roughly 2 to 5 times what the specialist GPU clouds charge. An A100 80GB is about $3.43/GPU-hr on AWS, $3.67 on Azure and $5.03 on Google Cloud, against $1.19-1.39 on RunPod and $1.99 on Lambda. That spread is the single largest cost lever in a self-hosted inference budget, and it is larger than any saving from picking a smaller model.
Three costs that are absent from every headline rate: egress, storage for the weights, and idle time. A 70B model at Q4 is tens of gigabytes that must be pulled to the instance on every cold start, and an endpoint billed hourly is billed while it sits idle waiting for a request. Rates read 2026-09-06; check each provider's live pricing on the directory hub.
Free tiers carrying this model
| Provider | Type | The catch | Live limits |
|---|---|---|---|
| Groq | Inference host | Free-plan limits are per-model and vary widely across the catalogue. Token-per-day caps bite before request caps on long prompts. | check → |
| OpenRouter | Aggregator | Free-pool membership changes without notice — a model you built on can stop being free, and the list above will drift. Data passes through OpenRouter and then the upstream provider, so check the privacy and routing settings before sending anything sensitive. | check → |
| Cerebras Inference | Inference host | Small model catalogue compared with the aggregators. You are choosing it for speed, not selection. | check → |
| NVIDIA NIM | Inference host | It is a credit grant, not a permanent free tier. When the credits run out you are on a paid plan or self-hosting the container. | check → |
Jurisdiction: China
This is the crucial split most write-ups miss: using a Chinese lab's HOSTED API sends your data to Chinese infrastructure and is a real procurement question. Downloading their OPEN WEIGHTS and running them on your own hardware, or on a Western host, sends nothing anywhere. DeepSeek and Z.ai publish under MIT; Alibaba publishes most of Qwen3 under Apache 2.0. Those are among the most permissive licences on this page.
Watch for: Several US states and a number of government bodies restrict Chinese-hosted AI services on official devices. That restriction is about the hosted service, not about the weights running in your own VPC — but expect to have to explain the difference to a security reviewer.
This model has a published technical report: arXiv:2505.09388 It is indexed in our research library alongside the work it builds on.
Frequently asked
Can I use Qwen3.8 27B commercially?
Apache 2.0, no conditions. Note that Alibaba did NOT extend Apache 2.0 to the whole family: Qwen3.8-Flash-Next and the 2.4T-A95B flagship both ship under a custom "other" licence. The 27B is the permissive one.
What hardware do I need to run Qwen3.8 27B?
24 GB GPU at Q4_K_M, or a 32 GB Mac at Q6_K. Weights alone are roughly 16 GB at Q4_K_M and 29 GB at Q8_0. Add KV cache on top of that, which grows with your context length.
What licence is Qwen3.8 27B released under?
Apache 2.0. Full commercial use, modification and redistribution. Patent grant included. The most permissive licence in common use for open-weight models.
Where can I use Qwen3.8 27B for free?
Free tiers carrying it include Groq, OpenRouter, Cerebras Inference, NVIDIA NIM. Limits differ per provider and change often, so check each provider's own limits page. You can also self-host the weights, which has no rate limit at all.
Similar models
DeepSeek-V4-Flash
Hosted, or 256 GB+ unified memory at Q4. Not a laptop model.
GLM-5.3-Flash
80-96 GB at Q4_K_M. A workstation or Ultra-class Mac.
Qwen3 Coder 30B A3B
24 GB GPU at Q4_K_M (18.6 GB), or a 32 GB Mac. Its 3.3B active parameters make it fast even on modest memory bandwidth.
Qwen3.6 27B
24 GB GPU at Q4_K_M, or a 24 GB Mac with modest context. Add about 2 GB for the MTP build.
Wiring Qwen3.8 27B into something real?
We build the evaluation harness, the failover and the cost ceilings around a model like this, so it survives contact with production.
Maps to AI agent development, AI automation development and AI strategy and consulting. Or see it working: our case studies.