Qwen·Text·Text Generation

Qwen3 235B

model = "qwen3-235b"

AvailableFree

Context128k tokens

Max output8,192 tokens

Providergonka

Slugqwen3-235b

Qwen3 235B — the default model on Gonka

Alibaba's 235-billion-parameter mixture-of-experts flagship. 22B active params per token — inference feels like a 70B dense model but generation quality is closer to 400B-class. Released mid-2025.

What it's good at
- Instruction-following, especially multi-step tasks
- Code generation and refactoring (Python, TS, Rust, Go)
- Summarization and extraction over long inputs (128k context)
- Tool-use and structured output (native JSON mode)

What to use something else for
- Creative writing with strong style control — bigger frontier models are still ahead
- Math Olympiad-level reasoning — wait for Kimi K2.5

Call it from a block

from openai import OpenAI
client = OpenAI(base_url="/v1", api_key="sk-run-...")  # auto-injected
resp = client.chat.completions.create(
    model="qwen3-235b",
    messages=[{"role": "user", "content": "hello"}],
)

Call it from a block

Every block gets a scoped key injected as OPENAI_API_KEY. Point your client at /v1.

from openai import OpenAI

client = OpenAI(
    base_url="/v1",
    api_key=os.environ["OPENAI_API_KEY"],  # injected by the runtime
)

resp = client.chat.completions.create(
    model="qwen3-235b",
    messages=[{"role": "user", "content": "hello"}],
)
print(resp.choices[0].message.content)

Build a block using this model

Blocks using Qwen3 235B

2 blocks call this model

@admin/hive-research

Hive Research

Agent

Autonomous deep-research agent. Give it a topic and a depth; it generates sub-questions, answers them, and synthesizes a polished markdown report.

Multi-agent swarm — a supervisor decomposes your task into sub-tasks, runs them in parallel with shell / file / web tools, then reconciles into a single answer. Model-agnostic; swap in Kimi K2.6 the day Gonka exposes it.

Job·qwen3-235bkimi-k2.5+1

32 runs