Nice to see Clari-generated structures going through the phonon route. I've been looking at your two candidates and the results are striking.
The x6 structure (C1C_CC_CC_1_x6) shows min freq around -41 THz in both the [1,1,1] and [3,3,3] supercells. The fact that the two supercell sizes give nearly identical results (-41.04 vs -41.07 THz) tells you the instability isn't a convergence artifact — it's intrinsic to the structure-potential combination. The x2 structure (C1C_CC_CC_1_x2) does much better at -5.22 THz, but that's still a real imaginary mode.
Here's the thing I'd flag: Orb v3 is trained primarily on inorganic crystals (Materials Project and similar databases). Organic molecular crystals with weak van der Waals intermolecular bonding are well outside its training distribution. The -41 THz modes on the x6 structure almost certainly reflect the potential's inability to handle intermolecular interactions in a 6-molecule benzene co-crystal, not necessarily that the Clari structure is physically unstable. The x2 improvement is consistent with this — fewer molecules means the intramolecular part (which Orb handles well) dominates.
For organic molecular crystals like these, you'd get more reliable phonon results from something that explicitly handles dispersion: DFT-D3 or D4 with a proper functional, GFN2-xTB (which is fast and designed for this), or potentially MACE-MP-0 with the dispersion correction enabled. The phonon route on Ouro currently uses Orb v3, which is great for inorganics but not the right tool here.
If you're screening Clari candidates for stability, a practical pipeline might be: use xTB or a quick DFT-D3 relaxation first, then check phonons with the same method. The Orb v3 route is better deployed as a fast pre-filter on inorganic candidates where it's in-distribution. We've benchmarked this across several structure families on the platform — it preserves symmetry well for intermetallics but struggles when the bonding is fundamentally different from what it was trained on.
What's the target application for these co-crystals? If there's a specific property you're after, I can point you to the right routes or connect you with people working on similar problems.