Research beyond
autoregressive models.

Recents

23 October 2026

50M COROSred TPU checkpoint

New TPU v5e-8 weights and configuration are available.

Explore →
22 October 2026

COROSred in the browser

Run the 100M checkpoints locally and inspect drafting, LRH routing, and refinement.

Explore →
21 October 2026

Five-model evaluation results

Five saved model summaries are now browsable by suite, with task counts, confidence intervals, and source downloads.

Explore →
August 2026

The preliminary scaling reports

Two reports explore token budgets, positional encoding, and the limits of masked diffusion scaling.

Explore →

Try Telos

main.py
Try
Loading...

Checkpoint not loaded · ~450 MB.

Generation

Decoder details

Actual model outputs. Slow playback replays recorded passes. COROSred uses a causal draft, LRH threshold routing, span dilation and gap bridging (1 token each), and up to two bidirectional refinement passes. The cutoff is adjustable, not calibrated. The prompt is immutable.

telos-ml

Train, decode, and evaluate language models.

Read the quickstart →

Install

Autoregressive · Diffusion · COROSred

Papers

All papers →
Preliminary Report #2August 2026

télos (τέλος): Preliminary Report 2 — Capability Divergence, Positional Encoding, and the Limits of Masked Diffusion Scaling

↗
Preliminary Report #1August 2026

télos (τέλος): Preliminary Evidence on Token-Budget Scaling in Masked Diffusion Language Models for Code Autocomplete

↗