Open weights stop 'competing' and start winning: Qwen 3 leads on coding, math, and long-context tasks
The framing has flipped. On coding, math, and long-context benchmarks, open-source models are no longer catching up to proprietary ones — they are winning, led by Qwen 3 235B-A22B under Apache 2.0. With DeepSeek V4, Kimi K2.6, and GLM closing the remaining gaps, the distance between a $200-a-month API bill and a self-hosted open model has never been smaller.
'Winning, not competing' is a real threshold, not a slogan. For years the open-weight story was 'nearly as good, much cheaper'; the claim now is that on specific high-value axes — coding, mathematical reasoning, long-context handling — the best open models top the charts outright. That changes the default: the open option becomes the first choice for those workloads, not the fallback.
Qwen 3 leading under Apache 2.0 is what makes the shift consequential rather than academic. A top-of-benchmark model that is also freely deployable seeds an ecosystem — Qwen has accounted for a large share of new model variants on Hugging Face — and every downstream project built on it is a reason the next one starts there too. Capability plus a permissive licence compounds.
The economic pressure this puts on the closed labs is exactly what the frontier price cuts reflect. When a self-hosted open model closes the capability gap and carries no per-token bill, the closed labs have to justify their price with either a capability lead or a cost cut — and this cycle they are reaching for both, cutting prices while racing to stay ahead on the benchmarks open weights are now topping.
Thunder Compute — Best open source LLMs (August 2026) → · Morph — The best open source LLMs (2026): ranked by benchmark, size, and use case →