OpenMined

We're building open-source tech that helps app builders & researchers get answers from data without direct access to it.
Join us on slack →

[OpenMined @openminedorg]

We're hiring a Head of Recruiting at OpenMined. We know the best candidates in AI and privacy aren't always looking. This role is for someone who finds them anyway and runs every pipeline from sourcing to offer.

Check out this role and others: https://t.co/4Dv9aRTcEg

[OpenMined @openminedorg]

That's two benchmarks in two separate domains where the top of the leaderboard is a fusion. We don't think it's the last.

[OpenMined @openminedorg]

Last month, fusions overtook every frontier model on DRACO.

[OpenMined @openminedorg]

Open-source AI vs. closed-source AI may be the wrong debate.

Our latest research points to model ensembles outperforming individual frontier models.

@iamtrask explains why

[OpenMined @openminedorg]

Costs, stated plainly: the fusion runs $0.296/task (aug_2026_trio in the figure attached). GPT-5.6-sol solo scores 58.1% at $0.056/task.

Solo models remain the efficiency play. Fusions buy accuracy that no single model reaches at any price. That capability only exists as a fusion.

[OpenMined @openminedorg]

That's two benchmarks in two separate domains where the top of the leaderboard is a fusion. We don't think it's the last.

[OpenMined @openminedorg]

Costs, stated plainly: the fusion runs $0.296/task (aug_2026_trio in the figure attached). GPT-5.6-sol solo scores 58.1% at $0.056/task.

Solo models remain the efficiency play. Fusions buy accuracy that no single model reaches at any price. That capability only exists as a fusion.

[OpenMined @openminedorg]

The reason hasn't changed since DRACO. Different models make mistakes in anti-correlated ways. Where one slips, another catches it. The newest frontier models don't outrun the network. They join it and make it stronger.

[OpenMined @openminedorg]

Best solo model under the same run, same judge, and same conditions: Claude Opus 5 at 58.3%.

[OpenMined @openminedorg]

The fusion: Kimi K3 + Claude Opus 5 + GPT-5.6-sol, with Opus 5 as synthesizer.

Score: 62.9% across all 525 HealthBench rows.

[OpenMined @openminedorg]

Last week we told you we hadn't built fusions with the new models yet...

Well, we just did and it set a new SOTA on HealthBench. 🧵

https://t.co/RbM5EQRtds

[OpenMined @openminedorg]

Last month, fusions overtook every frontier model on the DRACO deep research benchmark. New frontier models have shipped since, so we re-ran it with the same last-gen fusions vs. GPT-5.6 and Opus 5. The lead still holds. And we haven't even built fusions with the new models yet… https://t.co/NOcF0SLqmT

[OpenMined @openminedorg]

That's two benchmarks in two separate domains where the top of the leaderboard is a fusion. We don't think it's the last.

[OpenMined @openminedorg]

Costs, stated plainly: the fusion runs $0.296/task (aug_2026_trio in the figure attached). GPT-5.6-sol solo scores 58.1% at $0.056/task.

Solo models remain the efficiency play. Fusions buy accuracy that no single model reaches at any price. That capability only exists as a fusion.