Alibaba Qwen's 2.4-trillion-parameter MoE with 95B active parameters. Same card as Qwen3.8 Max; 1M-token context.
Qwen3.8-2.4T-A95B is Alibaba's open-weight Qwen3.8 checkpoint exposed in TheRouter as qwen/qwen3.8-2.4t-a95b. Public model cards describe it as a sparse Mixture-of-Experts text model with 2.4 trillion total parameters, 95 billion active parameters, a 1M-token context window, and Apache 2.0 weights published for independent serving.
This is not the same product surface as the proprietary Qwen3.8-Max API. The open-weight A95B route is text-only in TheRouter, while Max keeps the multimodal and vendor-operated service layer. Use the page to decide whether the open-weight economics and deployment control outweigh the missing vision and managed-platform features.
Treat it as a high-capacity text route for coding agents, long-document analysis, private deployment planning, and cost-sensitive fallback lanes. TheRouter's live catalog remains the operational source for price, context, output limits, modalities, and accepted request parameters; this curated layer adds citations and routing guidance only.
Alibaba publishes Qwen3.8-2.4T-A95B as an Apache 2.0 open-weight checkpoint and also lists a DashScope model ID with Qwen3.8 family serving behavior.
TheRouter serves qwen/qwen3.8-2.4t-a95b as a text-only OpenAI-compatible chat model. Catalog price, context, output cap, and request parameters come from standard-models.yaml rather than the open-weight model cards.
| Type | Rate |
|---|---|
| Input | $2.16 / 1M tokens |
| Output | $6.48 / 1M tokens |
| Architecture | Sparse Mixture-of-Experts with 2.4 trillion total parameters and 95 billion active parametersmodelscope.cn ↗ | verified |
| License | Apache 2.0huggingface.co ↗ | verified |
| Context window | 1,000,000 tokenshelp.aliyun.com ↗ | verified |
| Input and output | Text input and text output in TheRouter's cataloggithub.com/therouter/ai-router ↗ | verified |
| TheRouter max completion | 131,072 tokens in standard-models.yamlgithub.com/therouter/ai-router ↗ | verified |
| TheRouter list pricing | $2.16 input / $6.48 output per 1M tokens in the generated docs cataloggithub.com/therouter/ai-router ↗ | verified |
| Open-weight availability | Weights are published on Hugging Face and ModelScope for independent deploymenthuggingface.co ↗ | verified |
| Benchmark | Distribution | Score | Source |
|---|---|---|---|
Published A95B benchmark table The reviewed public model cards document architecture, context, license, and serving assumptions, but this no-budget pass did not find an official benchmark table dedicated to the open-weight A95B checkpoint. | — | Not publicly disclosed | — |
Qwen3.8-Max family benchmarks Alibaba reports Qwen3.8-Max benchmark scores for the managed Max service, but those should not be copied onto the open-weight A95B route without an explicit A95B benchmark ledger. | — | Not publicly disclosed | — |
TheRouter latency benchmark No live latency or throughput benchmark was run because this scheduled pass had no budget. The absence is recorded rather than filled with vendor or aggregator numbers. | — | Not publicly disclosed | — |
Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.
curl https://api.therouter.ai/v1/chat/completions -H "Content-Type: application/json" -H "Authorization: Bearer $THE_ROUTER_API_KEY" -d '{
"model": "qwen/qwen3.8-2.4t-a95b",
"messages": [
{"role": "user", "content": "Summarize the key points from this input."}
]
}'Use the standard OpenAI-compatible chat route with TheRouter's public model id for text-only requests.
curl https://api.therouter.ai/v1/chat/completions \
-H "Authorization: Bearer $THEROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"qwen/qwen3.8-2.4t-a95b","messages":[{"role":"user","content":"Audit this migration plan and list the three riskiest assumptions."}]}'TheRouter comparison explains where the open-weight checkpoint saves cost and where the managed Max route still owns multimodal and vendor-service workloads.
The managed Max launch gives the open-weight A95B page useful contrast because the two routes share family branding but not the same product capabilities.
The underlying Qwen3.8-2.4T-A95B checkpoint is published under Apache 2.0. The TheRouter model id is still an API route, so data residency, serving region, and operational behavior depend on TheRouter's configured upstream provider.
A95B is the open-weight text model route. qwen/qwen3.8-max is the managed Max API route with multimodal input and vendor-operated service features. Do not assume examples or benchmark claims for one transfer to the other without checking the source.
No. TheRouter's catalog lists this route as text input and text output. Send image, video, OCR, or screenshot tasks to qwen/qwen3.8-max or another configured vision model instead.
This scheduled pass had no budget and did not run snippet verification. Until an operator explicitly funds and approves that ledger run, the snippets component should remain capped rather than being inferred from unexecuted examples.
| source | URL | retrieved | |
|---|---|---|---|
| Architecture | modelscope.cn ↗ | 2026-09-21 | verified |
| License | huggingface.co ↗ | 2026-09-21 | verified |
| Context window | help.aliyun.com ↗ | 2026-09-21 | verified |
| Input and output | github.com/therouter/ai-router ↗ | 2026-09-21 | verified |
| TheRouter max completion | github.com/therouter/ai-router ↗ | 2026-09-21 | verified |
| TheRouter list pricing | github.com/therouter/ai-router ↗ | 2026-09-21 | verified |
| Open-weight availability | huggingface.co ↗ | 2026-09-21 | verified |
| Published A95B benchmark table | modelscope.cn ↗ | 2026-09-21 | unknown |
| Qwen3.8-Max family benchmarks | qwen.ai ↗ | 2026-09-21 | unknown |
| TheRouter latency benchmark | github.com/therouter/ai-router ↗ | 2026-09-21 | unknown |
| Qwen3.8 open-weight 2.4T-A95B vs Qwen3.8-Max: what changes for API routing | therouter.ai ↗ | 2026-09-21 | verified |
| Qwen3.8-Max becomes DashScope's top-tier model | therouter.ai ↗ | 2026-09-21 | verified |
| Is qwen/qwen3.8-2.4t-a95b open weight? | huggingface.co ↗ | 2026-09-21 | to verify |
| How is A95B different from qwen/qwen3.8-max? | therouter.ai ↗ | 2026-09-21 | to verify |
| Can I send images to qwen/qwen3.8-2.4t-a95b through TheRouter? | github.com/therouter/ai-router ↗ | 2026-09-21 | to verify |