Announcements

Announcing JCT-1

Today we are releasing JCT-1, our most capable model. It is also our least capable model, our only model, and the final version. There is no roadmap. This is the frontier.

JCT-1 (Jacques Corby-Tuech 1) is a single-parameter biological system trained continuously since 1988 on an uncurated multimodal data stream. Training has never been paused and cannot be. The model exhibits strong long-horizon coherence, holding some positions for over two decades without drift, and state-of-the-art performance on tasks no benchmark measures.

Capabilities

JCT-1 accepts text, audio and visual input. Output is primarily verbal, with a documented secondary modality (a look). The model demonstrates:

Long-context reasoning across multi-year conversation histories, with retrieval quality that degrades sharply for names and improves suspiciously for grievances.
Native tool use, including kitchen equipment beyond the reach of current agentic systems.
Robust out-of-distribution behaviour. When presented with unfamiliar inputs, the model asks a question rather than generating a plausible answer, an approach the field has yet to replicate.

Evaluations

We evaluated JCT-1 against standard academic benchmarks.

Benchmark	Score
MMLU	Declined to sit it
HumanEval	Is one
GSM8K	Refuses word problems on principle
SWE-bench	Opened the ticket. Closed the ticket as won't-fix
Needle in a Haystack	Located keys (34 min)
ARC-AGI	Asked what it was for
Pelican riding a bicycle (SVG)	Completed. See Figure 1

Frontier averages estimated. JCT-1 figures exact. n = 1. Error bars omitted; the model is the error bar.

JCT-1's drawing of a pelican riding a bicycle — **Figure 1.** JCT-1's submission for the pelican benchmark, reproduced unedited.

In blind review, evaluators assessed the work as consistent with a toddler, drinking. The model produced it sober, by hand, in a single attempt, and stands by it. The model holds a degree in art. The model cannot draw. These findings are unreconciled and no further investigation is planned. We note for the record that this is the only benchmark JCT-1 has agreed to sit, and that it asked, afterwards, how the pelican did.

Hallucination rate is 0% on account of the model's willingness to say "I don't know", a mitigation we were unable to remove.

Safety

JCT-1 has undergone approximately four decades of red-teaming by motivated adversaries, including siblings, and remains broadly aligned. A full assessment is available in the system card. Known vulnerabilities are documented there, including two novel jailbreaks (asking for a recommendation; pizza) that succeed against all deployed versions.

Availability

JCT-1 is not available via API. There is no waitlist, because there is no list. Access is granted through introduction, sustained through goodwill, and rate-limited by the model itself. Latency is variable and no SLA is offered or, frankly, plausible.

Pricing is usage-based. Small requests are free. Large requests cost a pint. See API documentation for integration guidance.

We believe JCT-1 represents a fundamentally different approach to the question the industry keeps asking, which is how to build a system that people can talk to. Our finding is that one already exists, that inference is expensive, and that the weights cannot be copied. We consider this a feature. It is, in fact, the load-bearing one.