Sleep Medicine Trends Q2 2026: Three Orexin Agonists, Compared Honestly

Twenty-three minutes. That is approximately the mean gain in maintenance-of-wakefulness sleep latency reported for the higher oveporexton doses in Takeda's phase 2b trial in narcolepsy type 1 — patients who entered the study falling asleep in roughly two minutes and left it staying awake through most of a 40-minute trial. It is the figure doing most of the work in every sleep medicine trends Q2 2026 briefing we have read, and on its own it is close to uninterpretable. Not wrong. Uninterpretable. Three orexin receptor agonists are now generating numbers in that neighborhood, and the numbers are not comparable to each other.

Why this class got a second life

The biology has been settled since 1999. Chemelli and colleagues (Cell, 1999) showed orexin knockout mice had cataplexy-like attacks; Lin and colleagues, in the same journal and year, traced canine narcolepsy to a mutation in the orexin-2 receptor gene. Nishino and colleagues (Lancet, 2000) found CSF hypocretin-1 undetectable in seven of nine narcolepsy patients. Thannickal and colleagues (Neuron, 2000) counted hypocretin neurons in postmortem brains — four narcolepsy cases against twelve controls, a small series that has never needed replacing — and found an 85 to 95 percent reduction.

So the target was never in doubt. The chemistry was. Takeda's first oral candidate, TAK-994, was halted in 2021 after hepatotoxicity — transaminase elevations serious enough to stop a program that was otherwise working. That failure is the reason the current generation is being watched as closely for liver enzymes as for wakefulness, and it is the single most important piece of context missing from most pipeline coverage.

How an orexin agonist produces wakefulness, in order

The drug is absorbed and crosses into the CNS. It binds OX2R on histaminergic neurons of the tuberomammillary nucleus — receptors that in narcolepsy type 1 are structurally intact but chronically unstimulated, because the neurons that should be releasing the peptide are gone. Those histamine neurons depolarize and fire. Histamine release rises across cortex and in the basal forebrain. Downstream, noradrenergic locus coeruleus and serotonergic raphe activity increases, partly through OX1R, partly through network effects.

The practical consequence is not stimulation in the amphetamine sense. It is stabilization. The sleep-wake switch in narcolepsy type 1 is a flip-flop circuit missing its stabilizing input, which is why these patients transition state so abruptly in both directions. Restoring receptor-level signaling restores the hysteresis. That is also the cleanest explanation for why the reported effects on cataplexy track the effects on sleepiness, rather than trailing them.

Three drugs, four criteria

The comparison worth making is not "which drug produced the biggest number." It is evidence depth, dosing behavior, indication breadth, and what has been observed on safety.

Read that table for asymmetry rather than for a winner. Oveporexton is the only one of the three whose pivotal-stage efficacy we can read in a journal rather than a slide deck. Alixorexton's advantage is not efficacy — nobody can honestly rank the two on effect size from public data — it is once-daily dosing and a broader indication net cast toward narcolepsy type 2 and idiopathic hypersomnia, populations that are not hypocretin-deficient and where the mechanism is being extended past the evidence that justifies it. ORX750 has produced striking early numbers in cohorts small enough that a single responder moves the mean.¹

Will orexin agonists help obstructive sleep apnea or ordinary sleepiness?

There is no evidence yet that they will, and the mechanism does not obviously predict it. Residual excessive daytime sleepiness in treated OSA is not a hypocretin-deficiency state. The orexin system is intact in these patients; the sleepiness comes from fragmentation, intermittent hypoxemia, and possibly wake-promoting neuron injury from years of the former two. Pushing an intact receptor system harder is a different proposition from replacing an absent ligand, and it is the proposition with the least data behind it.

Expect this to be tested anyway. The commercial logic is overwhelming — OSA with residual sleepiness is orders of magnitude larger than narcolepsy type 1, which affects roughly 25 to 50 per 100,000. Watch for those trials to be designed with Epworth as the primary endpoint rather than MWT, because subjective sleepiness scales are more forgiving in populations without a floor effect on objective latency.

What the MWT number hides

Mean sleep latency on the MWT is bounded. A 40-minute protocol ceilings at 40 minutes; a 20-minute protocol ceilings at 20. A patient who starts at two minutes has 38 minutes of headroom. A patient who starts at eight has 32. The mean change therefore encodes baseline severity and protocol length at least as much as drug potency, which makes cross-trial comparison of these headline figures a category error — one being committed enthusiastically right now, including by people who know better.

The more informative statistic, and the one to ask for when a sales team is in the room, is the proportion of patients reaching the ceiling, alongside the distribution rather than the mean. A drug that normalizes 70 percent of patients and does nothing for 30 percent produces the same average as one that half-fixes everyone. Those are different drugs.

What this does to the clinic

If an oral agent normalizes objective wakefulness in narcolepsy type 1, the bottleneck moves upstream to diagnosis. CSF hypocretin-1 below 110 pg/mL is the ICSD-3 criterion, and US assay access remains limited enough that most centers still route patients through MSLT. Reimbursement will follow the phenotype that has trial data, which means payers will likely demand documented hypocretin deficiency or a clean MSLT before authorizing a drug that will not be priced modestly. Expect prior-authorization friction of the kind sodium oxybate prescribers already know, and expect MSLT capacity to be the rate-limiting step in your schedule rather than the drug supply.

Rule of thumb: for any patient starting an orexin agonist inside or outside a trial, obtain baseline transaminases and repeat at four and eight weeks, and ask specifically about urinary frequency and blurred vision at each visit rather than waiting for spontaneous report. The hepatic monitoring is warranted by TAK-994's history, not by anything the current compounds have done; the on-target questions are warranted by what they have.

The verdict

OX2R agonism reversing sleepiness in hypocretin-deficient narcolepsy is now well-established — replicated across species, mechanism-consistent, and confirmed in randomized human trials with peer-reviewed publication. Extension to narcolepsy type 2 and idiopathic hypersomnia is plausible but thin: real trials are running, no pivotal readouts are in hand, and the biological rationale is weaker than the enthusiasm. Extension to residual sleepiness in OSA is, for now, folk wisdom dressed in mechanism language.

Which returns us to twenty-three minutes. It is a real number from a real trial in a well-defined population, and it says almost nothing about the two drugs behind it, the three indications being claimed for the class, or what any of this will cost the patient in your clinic next year. The number is the beginning of the question, not the answer to it.