Model
Qwen3.8 2.4T A95B
Recorded as suitable for chat, rag, coding, reasoning, multilingual, zh, agentic.
- Source
- Model registry API
- Verified
- 19 Sep 2026
- Confidence
- High
01What this is
- Author
- Qwen
- Family
- Qwen3.8
- Task
- text-generation
- Hugging Face
- Qwen/Qwen3.8-2.4T-A95B
- OpenRouter
- qwen/qwen3.8-2.4t-a95b
- Also known as
- Qwen3.8 2.4T, qwen3.8-2.4t-a95b, qwen 2.4t
02Registry facts
- LicenceFACT
other
- FreshRetrieved 19 Sep 2026
- ParametersFACT
2446.2B
- FreshRetrieved 19 Sep 2026
- Context lengthFACT
not returned by the registry
- Evidence not verified
- Downloads (30 days)FACT
56,249
- FreshRetrieved 19 Sep 2026
- LikesFACT
1,233
- FreshRetrieved 19 Sep 2026
- GatedFACT
no
- FreshRetrieved 19 Sep 2026
Model
- downloadsFACT
56249
- Verification may be outdatedRetrieved 19 Sep 2026
- gatedFACT
false
- Verification may be outdatedRetrieved 19 Sep 2026
- hosted context length · SiliconFlowFACT
1048576
- Verification may be outdatedRetrieved 19 Sep 2026
- hosted context length · AlibabaFACT
1000000
- Verification may be outdatedRetrieved 19 Sep 2026
- hosted context length · NovitaFACT
1000000
- Verification may be outdatedRetrieved 19 Sep 2026
- likesFACT
1233
- Verification may be outdatedRetrieved 19 Sep 2026
- model licenseFACT
other
- Verification may be outdatedRetrieved 19 Sep 2026
- parameter countFACT
2446182725504
- Verification may be outdatedRetrieved 19 Sep 2026
Pricing
- hosted price completion per m · SiliconFlowFACT
6.0000
- AgeingRetrieved 19 Sep 2026
- hosted price completion per m · AlibabaFACT
6.0000
- AgeingRetrieved 19 Sep 2026
- hosted price completion per m · NovitaFACT
6.0000
- AgeingRetrieved 19 Sep 2026
- hosted price prompt per m · SiliconFlowFACT
2.0000
- AgeingRetrieved 19 Sep 2026
- hosted price prompt per m · AlibabaFACT
2.0000
- AgeingRetrieved 19 Sep 2026
- hosted price prompt per m · NovitaFACT
2.0000
- AgeingRetrieved 19 Sep 2026
Other
- available on platform · alibabaFACT
alibaba
Evidence not verified — the quoted phrase was not found in the fetched document.
- Evidence not verified
- last modifiedFACT
2026-08-12T10:24:04.000Z
- Verification may be outdatedRetrieved 19 Sep 2026
- libraryFACT
transformers
- AgeingRetrieved 19 Sep 2026
03VRAM by quantisation
Peak VRAM for serving this model, computed by lib/deployment/hardware.ts from weights + KV cache + 15% runtime overhead. Inputs: 2446.2B parameters, 8,192 tokens of context (assumed — none fetched yet), 4 concurrent requests (assumed).
| Quantisation | Weights | Total |
|---|---|---|
| bf16bf16 is two bytes per weight, the same arithmetic as fp16. | 4892.37 GB | 5643.51 GB |
| fp8 | 2446.18 GB | 2830.4 GB |
| int4 | 1223.09 GB | 1423.84 GB |
What this estimate assumes
- Weights: 2446.2B parameters × 2 bytes per parameter (fp16) = 4892.37 GB.
- No model configuration was supplied, so the KV cache uses the size heuristic for this band: 0.459 MB per token, from extrapolated above 90B — no configuration was fetched for this band.
- KV cache: 0.459 MB per token × 8,192 tokens of context × 4 concurrent requests = 15.03 GB, sized for every user holding a full context at once.
- Overhead: 15% of weights plus KV cache for the runtime, activations and memory fragmentation = 736.11 GB.
- GB means 10⁹ bytes, matching how GPU memory is advertised.
- KV cache held at fp16 (2 bytes per element); quantising weights does not by itself quantise the cache.
- Peak figure, not average: prefix caching and paged attention usually keep real usage lower, while speculative decoding, CUDA graphs and a second resident model push it higher.
- "Concurrent" counts requests generated at the same instant, not people: 4 simultaneous requests is the worst case this estimate is built on — measure yours before buying.
04Hardware profiles
No hardware profile was sized for this answer.
05Hosted providers
- Prompt (per million tokens, USD)FACT
2
Source: openrouter
- FreshRetrieved 19 Sep 2026
- Completion (per million tokens, USD)FACT
6
- FreshRetrieved 19 Sep 2026
| Provider | Context |
|---|---|
| Novita | 1,000,000 |
| Alibaba | 1,000,000 |
| SiliconFlow | 1,048,576 |
| Venice | 262,144 |
| Modal | 1,000,000 |
| DeepInfra | 262,144 |
06Deployment stacks
No published deployment stack names this model yet.
07Evidence
- 01–09Tier 4official repository / model cardOpenRouter
OpenRouter endpoints — qwen/qwen3.8-2.4t-a95b
https://openrouter.ai/api/v1/models/qwen/qwen3.8-2.4t-a95b/endpoints
Verification may be outdatedRetrieved 19 Sep 2026sha256:9e61292c5ab29 records · · · · · · · ·
- 10–16Tier 4official repository / model cardHugging Face Hub
Hugging Face model card data — Qwen/Qwen3.8-2.4T-A95B
https://huggingface.co/api/models/Qwen/Qwen3.8-2.4T-A95B
Verification may be outdatedRetrieved 19 Sep 2026sha256:9c247922baa97 records · · · · · ·
Improve this page
Sign in to contribute
From the field
0 deployments · 0 questions
Nobody has reported deploying this here yet, and no question has been opened against this page. Both appear once a reviewer accepts them.