OpenAI hasn’t launched an open-weight language mannequin since GPT-2 again in 2019. Six years later, they stunned everybody with two: gpt-oss-120b and the smaller gpt-oss-20b.
Naturally, we needed to know — how do they really carry out?
To seek out out, we ran each fashions via our open-source workflow optimization framework, syftr. It evaluates fashions throughout completely different configurations — quick vs. low cost, excessive vs. low accuracy — and consists of help for OpenAI’s new “pondering effort” setting.
In concept, extra pondering ought to imply higher solutions. In observe? Not all the time.
We additionally use syftr to discover questions like “is LLM-as-a-Judge actually working?” and “what workflows perform well across many datasets?”.
Our first outcomes with GPT-OSS would possibly shock you: one of the best performer wasn’t the most important mannequin or the deepest thinker.
As a substitute, the 20b mannequin with low pondering effort persistently landed on the Pareto frontier, even rivaling the 120b medium configuration on benchmarks like FinanceBench, HotpotQA, and MultihopRAG. In the meantime, excessive pondering effort not often mattered in any respect.
How we arrange our experiments
We didn’t simply pit GPT-OSS in opposition to itself. As a substitute, we needed to see the way it stacked up in opposition to different robust open-weight fashions. So we in contrast gpt-oss-20b and gpt-oss-120b with:
- qwen3-235b-a22b
- glm-4.5-air
- nemotron-super-49b
- qwen3-30b-a3b
- gemma3-27b-it
- phi-4-multimodal-instruct
To check OpenAI’s new “pondering effort” characteristic, we ran every GPT-OSS mannequin in three modes: low, medium, and excessive pondering effort. That gave us six configurations in whole:
- gpt-oss-120b-low / -medium / -high
- gpt-oss-20b-low / -medium / -high
For analysis, we forged a large web: 5 RAG and agent modes, 16 embedding fashions, and a variety of movement configuration choices. To guage mannequin responses, we used GPT-4o-mini and in contrast solutions in opposition to identified floor fact.
Lastly, we examined throughout 4 datasets:
- FinanceBench (monetary reasoning)
- HotpotQA (multi-hop QA)
- MultihopRAG (retrieval-augmented reasoning)
- PhantomWiki (artificial Q&A pairs)
We optimized workflows twice: as soon as for accuracy + latency, and as soon as for accuracy + value—capturing the tradeoffs that matter most in real-world deployments.
Optimizing for latency, value, and accuracy
Once we optimized the GPT-OSS fashions, we checked out two tradeoffs: accuracy vs. latency and accuracy vs. value. The outcomes had been extra stunning than we anticipated:
- GPT-OSS 20b (low pondering effort):
Quick, cheap, and persistently correct. This setup appeared on the Pareto frontier repeatedly, making it one of the best default alternative for many non-scientific duties. In observe, meaning faster responses and decrease payments in comparison with greater pondering efforts. - GPT-OSS 120b (medium pondering effort):
Greatest suited to duties that demand deeper reasoning, like monetary benchmarks. Use this when accuracy on complicated issues issues greater than value. - GPT-OSS 120b (excessive pondering effort):
Costly and often pointless. Maintain it in your again pocket for edge instances the place different fashions fall quick. For our benchmarks, it didn’t add worth.
Studying the outcomes extra rigorously
At first look, the outcomes look simple. However there’s an essential nuance: an LLM’s prime accuracy rating relies upon not simply on the mannequin itself, however on how the optimizer weighs it in opposition to different fashions within the combine. For instance, let’s take a look at FinanceBench.
When optimizing for latency, all GPT-OSS fashions (besides excessive pondering effort) landed with related Pareto-frontiers. On this case, the optimizer had little motive to focus on the 20b low pondering configuration—its prime accuracy was solely 51%.
When optimizing for value, the image shifts dramatically. The identical 20b low pondering configuration jumps to 57% accuracy, whereas the 120b medium configuration truly drops 22%. Why? As a result of the 20b mannequin is much cheaper, so the optimizer shifts extra weight towards it.
The takeaway: Efficiency is determined by context. Optimizers will favor completely different fashions relying on whether or not you’re prioritizing velocity, value, or accuracy. And given the massive search house of doable configurations, there could also be even higher setups past those we examined.
Discovering agentic workflows that work effectively in your setup
The brand new GPT-OSS fashions carried out strongly in our assessments — particularly the 20b with low pondering effort, which frequently outpaced dearer rivals. The larger lesson? Extra mannequin and extra effort doesn’t all the time imply extra accuracy. Typically, paying extra simply will get you much less.
That is precisely why we constructed syftr and made it open-source. Each use case is completely different, and one of the best workflow for you is determined by the tradeoffs you care about most. Need decrease prices? Sooner responses? Most accuracy?
Run your own experiments and discover the Pareto candy spot that balances these priorities in your setup.
