Qwen
last releaseApril 29, 2025
powered byQwen 3, Qwen 3.5, Qwen3-Coder-Next
goblin vibe check:
open-source llm that's surprisingly good at code and multilingual tasks without needing a phd to run it
alibaba's open model family for coding, multilingual chat, reasoning, and agent workflows with strong local deployment options across a wide size range.
context
262k
tokens
cost
$0.30
per $1.20
Broad open model family spanning lightweight local builds through larger coding and agent modelsStrong multilingual coding and reasoning with competitive open benchmark performanceNative long-context support reaching roughly 262K tokens on newer family variantsOpen weights and active local ecosystem across Ollama, Hugging Face, vLLM, and llama.cpp-adjacent tooling
key features
Broad open model family spanning lightweight local builds through larger coding and agent modelsStrong multilingual coding and reasoning with competitive open benchmark performanceNative long-context support reaching roughly 262K tokens on newer family variantsOpen weights and active local ecosystem across Ollama, Hugging Face, vLLM, and llama.cpp-adjacent toolingflagship and small MoE variants both target strong coding and reasoning
spec & usage
Best fit for local coding assistants, self-hosted chat, multilingual copilots, and agent backends that need open weights
Deployable through standard inference stacks with both API-hosted and self-hosted workflows available
contexts range from 32K on the smallest checkpoints to 128K on larger releases
limitations
Best coding and reasoning quality still comes from larger checkpoints that are heavier for consumer hardware
Open family performance varies more by checkpoint choice than tightly packaged commercial products
scope:
codelanguageagentresearchlocalcloudopen-sourcebenchmark-strongfreemultimodal
launchAugust 3, 2023
last releaseApril 29, 2025