Phase 3, finally: what the readouts actually say

For years the honest answer to "does psilocybin therapy work?" was "the early trials look remarkable, and early trials often do." That answer has now changed, though not as simply as the headlines suggest.

Where the evidence stood

The landmark phase 2b result was published in the New England Journal of Medicine in November 2022 (Goodwin et al.). It enrolled 233 people with treatment-resistant depression, randomised across three doses of COMP360 psilocybin: 25mg, 10mg, and 1mg as an active control. Depression was measured on the MADRS scale.

At week 3, the 25mg group improved by 12.0 points against 5.4 for the 1mg control, a difference of 6.6 points. The 10mg dose did not separate from control at all. Two details from that paper get quietly dropped in most coverage. Sustained response at week 12 was not supportive of the headline result, at 20.3% against 10.1%. And suicidal ideation or self-injury occurred in all three dose groups, including three participants in the 25mg arm who did not respond to treatment.

What phase 3 has reported

Two readouts, both from Compass Pathways, both announced by press release rather than published in a journal. That distinction matters and we will come back to it.

COMP005, announced June 2025, dosed 258 participants across 32 US sites, comparing a single 25mg dose against placebo. The difference on MADRS at week 6 was 3.6 points in psilocybin's favour, with a p-value below 0.001. It was the first phase 3 efficacy readout for any classic psychedelic.

COMP006, announced February 2026, was larger: 581 participants dosed, two sessions three weeks apart, across North America and Europe. The 25mg arm beat the 1mg comparator by 3.8 MADRS points at week 6. A 26-week durability analysis followed in July 2026.

The number that should temper the excitement

Phase 3 roughly halved the effect. From 6.6 points in phase 2b to 3.6 and 3.8. This is the most ordinary story in clinical research: effects shrink as trials get bigger, more rigorous and more representative. It does not mean the earlier result was wrong. It means the honest estimate of how much this helps, on average, is more modest than the first wave of coverage implied.

Also worth knowing: the durability figures released in 2026 use a threshold of at least 25% improvement on MADRS to define a meaningful response. The conventional bar in depression research, and the one used in Compass's own phase 2b, is 50%. Read any percentage in this field alongside the definition it uses.

The blinding problem nobody has solved

You cannot blind a psychedelic. Participants know whether they received 25mg of psilocybin or an inert capsule, usually within the hour, and so do the therapists sitting with them. That is called functional unblinding, and it means expectancy is baked into every result.

This is not a fringe objection. When the FDA's advisory committee reviewed MDMA-assisted therapy for PTSD in June 2024, it voted 9 to 2 against effectiveness and 10 to 1 against a favourable benefit-risk profile, and the committee minutes cite functional unblinding and unmanaged expectation bias as central to why. The application was rejected that August. Psilocybin faces the same methodological difficulty.

So where does that leave things

A fair summary, as of mid-2026: psilocybin-assisted therapy has now shown a statistically robust benefit in large, controlled, multi-site trials for a condition where existing treatments have failed. The benefit is real and modest. The safety picture requires care rather than reassurance. The phase 3 data has not yet been through peer review, and until it is, every figure above should be read as what a company chose to announce about its own drug.

That is still a genuinely hopeful position, and it is a more useful one than either the miracle framing or the backlash. The research that gets us to a firmer answer is the research a share of every sale here helps fund.