Open weights from Alibaba

Qwen3.6 online, pay per use

Alibaba's Qwen team shipped Qwen3.6 in April 2026 under an Apache 2.0 license: a dense 27B flagship and an efficient 35B-A3B mixture of experts with only 3B active parameters. It reads up to about a million tokens, accepts text, images and video, and its 27B beats far larger models on agentic coding at 77.2 percent on SWE-bench Verified. On avots.ai you run every Qwen3.6 tier pay per use, no subscription, next to GPT-5.6, Claude Sonnet 5, Gemini and 40+ media models.

Use Qwen3.6 now ✦ See pricing
The tiers

Qwen3.6-27B, 35B-A3B and Flash: which to pick

 Qwen3.6-27BQwen3.6-35B-A3BQwen3.6-Flash
Typedense 27B flagshipMoE, 3B activefast hosted tier
LicenseApache 2.0, openApache 2.0, openhosted API
Context window256K, ~1M via YaRN256K, ~1M via YaRN1M tokens
Inputtext, image, videotext, image, videotext, image, video
On avots.aipay per usepay per usepay per use

Pick the dense 27B for the strongest single answers, the 35B-A3B MoE when you want frontier quality at a low active cost, and Flash or Plus when speed and a full million tokens matter more than depth. Switch freely, one balance covers them all.

What is new

What Qwen3.6 brings

Beats bigger models

Qwen3.6-27B scores 77.2% on SWE-bench Verified and 87.8% on GPQA Diamond, outrunning the larger Qwen3.5-397B MoE on agentic coding.

Open weights, Apache 2.0

Download and self-host commercially. The 27B runs in about 18GB VRAM at 4-bit, or run it hosted on avots.ai with no GPU.

Up to 1M context, multimodal

262,144 native tokens, extensible to about a million via YaRN, with text, image and video input across both open variants.

Hybrid attention and thinking

Gated DeltaNet plus gated self-attention, with a hybrid thinking mode on by default and Thinking Preservation across a chat.

Three steps

How to use Qwen3.6

Create an account

Sign up on app.avots.ai with email, Google or Telegram. Takes a minute.

Top up from 0.99 EUR

1,000 tokens cost 0.99 EUR. No plan to choose, no renewal, the balance sits there until you use it.

Pick a Qwen3.6 tier

Choose Qwen3.6-27B, 35B-A3B, Plus or Flash in the model picker, or let Agent Avots route. Every reply shows its token cost.

Sign up and start
One balance

Where you can use Qwen3.6

Web app

app.avots.ai on desktop and mobile, no install needed.

Telegram

Qwen3.6 in @AvotsAIbot, same balance, from any phone.

MCP

Use it inside Claude Desktop, Cursor or Cline via the MCP server.

Try Qwen3.6 for under a euro

1,000 tokens cost 0.99 EUR and a Qwen3.6 reply costs a handful of them. The same balance also unlocks GPT-5.6, Claude Sonnet 5, Gemini, Kimi K3, Veo video and Nano Banana images.

Sign up and chat ✦
Guides

Related pages

Kimi K3

Moonshot's frontier open model, number one at coding, 1M context.

DeepSeek V4

DeepSeek's open frontier model, huge context, very low price.

GLM-5.2

Z.ai's open-weight leader with a 744B MoE and top GPQA scores.

Claude Sonnet 5

Anthropic's most agentic Sonnet, 1M context, near Opus quality.

avots.ai runs 40+ AI models for chat, image, video and audio on one balance. See the full list on the Tools page or open the app and start creating.

Browse all model pages Open the app
FAQ

Qwen3.6 questions

What is Qwen3.6?
Qwen3.6 is Alibaba's April 2026 open-weight LLM generation, released under an Apache 2.0 license. It ships as a dense 27B flagship and an efficient 35B-A3B mixture of experts with only 3B active parameters. Both read up to about a million tokens and accept text, images and video.
How much does Qwen3.6 cost on avots.ai?
Pay per use, no subscription. Per million tokens the 35B-A3B is about 0.14 input and 1.00 output, the 27B about 0.45 and 2.70, the hosted Plus about 0.33 and 1.95, and the Max about 1.04 and 6.24. Top up from 0.99 EUR and every reply shows its token cost.
What is Qwen3.6 best at?
Agentic coding, math and reasoning. Qwen3.6-27B scores 77.2 percent on SWE-bench Verified, 87.8 percent on GPQA Diamond, 94.1 percent on AIME 2026 and 86.2 percent on MMLU-Pro, beating far larger models on several tasks.
Is Qwen3.6 really open source?
Yes. Qwen3.6 is released under Apache 2.0, so you can download the weights and self-host commercially. The 27B runs in about 18GB VRAM at 4-bit quantization. On avots.ai you can also run it hosted with no GPU, pay per use.
Can I use Qwen3.6 with GPT-5.6 and Claude on one balance?
Yes. One balance covers 40+ models: Qwen3.6, GPT-5.6, Claude Sonnet 5, Gemini and Kimi K3 for chat, plus Veo, Sora and Nano Banana for media. Use them in the web app, Telegram, MCP or the API.
Try it on avots ✦
Start creating