Model Details
About
Qwen 3.6 35B Thinking (35B MoE (3B active)) is a chat model built by Alibaba Cloud. It accepts up to 32K tokens of context per request. Strongest at Reasoning, math, step-by-step thinking. Self-hosted on Free.ai GPUs — runs free against your daily token pool (100 tokens per message). Released under Apache 2.0 — commercial use permitted on Free.ai.
Use via API
curl https://api.free.ai/v1/chat/ \
-H "Authorization: Bearer YOUR_KEY" \
-d '{"model":"deepseek-r1-7b"}'
Compare
FAQ
DeepSeek-R1 7B is no longer served on Free.ai. Requests for it are now answered by Qwen 3.6 35B with thinking mode on, our self-hosted chat model, which reasons step by step before it answers.
Qwen 3.6 35B Thinking (35B MoE (3B active)) is a chat model built by Alibaba Cloud. It accepts up to 32K tokens of context per request. Strongest at Reasoning, math, step-by-step thinking. Self-hosted on Free.ai GPUs — runs free against your daily token pool (100 tokens per messa
Qwen 3.6 35B Thinking works well for Reasoning, math, step-by-step thinking. Try the sample prompts above to see its style.
About 100 tokens per average message. Free accounts get a 30,000-token daily pool that covers self-hosted chat; premium models are pay-as-you-go, with token top-ups from $1.
It depends on the task. /chat/compare/ lets you send the same prompt to Qwen 3.6 35B Thinking and any other model side-by-side - comparison is the fastest way to decide.
Yes. Outputs are yours - Free.ai does not claim rights to anything you generate. The underlying model is Apache 2.0-licensed.
32,768 tokens.
Replies stream token-by-token within ~1 second. Total response time depends on length and model size - small models stream faster, frontier models trade speed for depth.
Yes. Signed-in users see every chat in /account/?tab=history. You can also share a one-link copy of any conversation via the Share button.
Free.ai does not train models on your conversations. Self-hosted models stay on our GPUs. Premium models route to the upstream provider for inference.
Yes. POST to /v1/chat/ with model="deepseek-r1-7b" and a messages array. Streaming SSE is supported. Full reference: /api/.
Qwen 3.6 35B Thinking is Apache 2.0-licensed with 35B MoE (3B active) parameters. See /models/deepseek-r1-7b/ for setup notes and our open-source repos at github.com/freeaigit.
DeepSeek-R1 7B is retired: chat with Qwen 3.6 35B Thinking instead
DeepSeek-R1 7B is no longer served on Free.ai. Requests for it are now answered by Qwen 3.6 35B with thinking mode on, our self-hosted chat model, which reasons step by step before it answers.
What is Qwen 3.6 35B Thinking?
Qwen 3.6 35B Thinking (35B MoE (3B active)) is a chat model built by Alibaba Cloud. It accepts up to 32K tokens of context per request. Strongest at Reasoning, math, step-by-step thinking. Self-hosted on Free.ai GPUs — runs free against your daily token pool (100 tokens per message). Released under Apache 2.0 — commercial use permitted on Free.ai.
Best for: Reasoning, math, step-by-step thinking
Why use Qwen 3.6 35B Thinking for chat?
Streaming responses
Replies stream token-by-token within ~1 second of pressing Send. No idle waiting.
Saved history
Signed-in users see every chat in /account/?tab=history with one-click share links.
Compare side by side
Send the same prompt to up to 4 models at /chat/compare/ and judge the outputs side by side.
Commercial use OK
Outputs are yours. Use them in apps, ads, docs, or anything else without attribution.
Sample prompts
Pricing
Self-hosted on our GPUs. Generation draws from your daily free pool first; once that runs out, paid tokens start at $1. Roughly ~100 tokens per message.
Compare to alternatives
See all chat models → · Compare up to 4 chat models side-by-side →