benchr
Let's be exact about "free," because the word does a lot of work here. We're not talking about a free trial or a generous tier of a paid product. These are open-weight models. The weights are published under open licenses on Hugging Face, so you can download them and run them on your own hardware for nothing but compute. And each maker also runs a free hosted chat, so you can just use the model directly. That's a different thing from a paid IDE assistant, which wraps a model in tooling and bills you monthly. For that comparison, see the coding assistants shootout.
Three families lead the open-weight coding pack right now: DeepSeek out of China, Alibaba's Qwen, and Moonshot AI's Kimi. Here's how they stack up, by the scores the makers themselves publish.
| Model | Params (total / active) | License | Context | SWE-bench Verified |
|---|---|---|---|---|
| Kimi K2.6 | 1T / 32B | Modified MIT | 256K | 80.2 |
| DeepSeek V4-Pro | 1.6T / 49B | MIT | 1M | not published |
| Qwen3.6-35B-A3B | 35B / 3B | Apache 2.0 | 262K | 73.4% |
| Qwen3-Coder-Next | 80B / 3B | Apache 2.0 | 256K | ~71% |
Kimi K2.6: the score leader
If you go by the strongest officially published number, Moonshot AI's Kimi K2.6 wins. It posts 80.2 on SWE-bench Verified, the benchmark that scores real GitHub issue resolution, plus 58.6 on the tougher SWE-bench Pro and 76.7 on the multilingual variant. Those are frontier-class numbers from an openly licensed model, which a year ago would have sounded absurd.
Under the hood it's a 1-trillion-parameter Mixture-of-Experts model with 32 billion active per token, released in late April 2026 under a Modified MIT license. Moonshot leans hard into long-horizon agent work: K2.6 can coordinate up to 300 sub-agents across as many as 4,000 steps in a single run, with demonstrated 12-hour autonomous sessions. The catch is size. A trillion-parameter model is not something you spin up on a laptop, so for most people "free" here means the hosted chat or rented GPUs, not local. The full Kimi K2.6 review goes deeper on the Agent Swarm and what self-hosting actually costs.
Qwen: the one you can run yourself
Qwen's pitch is efficiency, and it's the reason it might be the right answer even though it doesn't top the table. The newest entry, Qwen3.6-35B-A3B from April 2026, scores 73.4% on SWE-bench Verified while activating just 3 billion parameters per token out of 35 billion total. Alibaba also ships a dedicated coder, Qwen3-Coder-Next, an 80B model with the same 3B active footprint that lands a bit over 70% depending on the scaffold. Both are Apache 2.0, the most permissive license here, with no strings on commercial use.
That tiny active-parameter count is the headline. A 3B-active model runs on hardware that a trillion-parameter model can't touch, which puts a strong coder within reach of a single good GPU. If "free" has to mean "free on my machine," Qwen is the family to start with, and the small language models piece explains why the active-parameter trick works.
DeepSeek V4: the long-context free chat
DeepSeek V4 landed in April 2026 in two MIT-licensed variants, V4-Pro at 1.6 trillion parameters and a lighter V4-Flash, both with a full 1-million-token context window, the longest in this group. DeepSeek positions V4 as open-source state of the art on agentic coding. Here's the honest caveat: DeepSeek has not published a specific SWE-bench Verified score for V4, and the third-party numbers making the rounds contradict each other, so we won't print one. The last DeepSeek model with an official figure is V3.2, at 73.1 on SWE-bench Verified.
What DeepSeek does have is the smoothest free path. You can use V4 right now at chat.deepseek.com at no cost, with that million-token window intact, which makes it the easy choice for chewing through a large codebase in one session even before you download anything. Where it sits against the rest of the open field is covered in the open-weight tier right now.
So which one
Go with Kimi K2.6 if you want the strongest published coder and you can host it or use the hosted chat. Go with Qwen, specifically Qwen3.6-35B-A3B, if you need to run the model yourself on reasonable hardware; it's the best score-per-gigabyte here by a distance. Reach for DeepSeek V4 when you want the longest context and a zero-setup free chat, and treat its coding strength as strong-but-unscored until DeepSeek publishes a number.
One reality check before you commit: the two leaders are trillion-parameter models, so "free to download" and "free to run" are not the same sentence. If self-hosting is the goal, read running models on your own machine first, then size your ambitions to your GPUs.