Implementations

What you can actually run. Repository statistics verified September 2026.

Latent reasoning is research code, with one exception. The method-specific repositories are self-contained academic releases at GPT-2 to 8B scale, without packaging and frequently pinned to a stale dependency snapshot. Most reproduce their paper's numbers. None is something to add to a serving stack.

Open model checkpoints

Two families of open weights exist that reason by depth recurrence.

ModelScaleNotesLicense
Ouro 1.4B / 2.6B
7.7T tokens
Looped language model, 24 layers, 4 recurrent steps by default with configurable early exit. Instruction-tuned -Thinking variants published alongside. Loads through standard Hugging Face tooling. Claims 3–4B dense-equivalent quality at 2.6B parameters. The only release here a practitioner could plausibly deploy rather than study. Apache-2.0
Huginn-0125 3.5B
800B tokens
Prelude / recurrent core / coda, with user-controlled test-time recursion (num_steps, typically 4–64) and entropy, KL and argmax-stability early-exit criteria. Eight intermediate training checkpoints also published. Explicitly a proof of concept and undertrained relative to modern dense models of the same size. Apache-2.0

Method implementations

RepositoryMethodStars / last pushState
facebookresearch/coconut Coconut: continuous thoughts
arXiv:2412.06769
1,705 · 2026-07 Training code and data for the GPT-2-scale GSM8K / ProsQA / ProntoQA experiments. No official checkpoints; community reproductions exist on the Hub. MIT.
seal-rg/recurrent-pretraining Recurrent depth (Huginn)
arXiv:2502.05171
961 · 2025-12 Full pretraining stack behind Huginn-0125. Well documented, actively maintained. Apache-2.0.
raymin0223/mixture_of_recursions Mixture-of-Recursions
arXiv:2507.10524
626 · 2025-09 Complete pretraining framework with per-token depth routing and matched KV caching. The most production-adjacent of the architecture-level methods. Apache-2.0.
ezelikman/quiet-star Quiet-STaR
arXiv:2403.09629
740 · 2024-08 Stale. Patches a specific transformers dev snapshot from 2024 into Mistral's modeling code; brittle to run against current versions. Apache-2.0.
InternLM/SIM-CoT Supervised implicit CoT
arXiv:2509.20317
215 · 2026-04 The best-resourced and most current implicit-CoT repository. Reports gains over CoT, Coconut and CODI on GPT-2 and Llama 1B/3B/8B. Apache-2.0.
da03/Internalize_CoT_Step_by_Step Stepwise internalisation
arXiv:2405.14838
208 · 2025-04 Ships data (including GSM8K-Aug), pretrained checkpoints and eval scripts. Genuinely reproduces the paper. MIT.
zjunlp/LightThinker Gist-token compression
arXiv:2502.15589
165 · 2026-06 Actively maintained. MIT.
zhenyi4/codi CODI: self-distillation
arXiv:2502.21074
107 · 2025-12 Self-contained GPT-2-scale training and eval; reproduces the GSM8K numbers. No license file: check before reuse.
xuyige/SoftCoT SoftCoT and SoftCoT++
arXiv:2502.12134, 2505.11484
95 · 2025-05 Training scripts, GSM8K preprocessing, checkpoints. No explicit license despite being public: check before reuse.
JacobPfau/fillerTokens Filler tokens
arXiv:2404.15758
76 · 2024-04 Minimal research code for two synthetic algorithmic tasks. Dormant. No license.
mcleish7/retrofitting-recurrence Retrofitted recurrence
arXiv:2511.07384
69 · 2025-11 Converts a pretrained non-recurrent model to depth-recurrent via curriculum. Apache-2.0.
yingfan-bot/lotus LOTUS: supervised loop latents
arXiv:2606.31779
51 · 2026-07 Fine-tuning scripts over Llama-3.2-3B. no standalone checkpoint. MIT.
alperengozeten/CoT2 Continuous CoT / parallel search
arXiv:2505.23648
16 · 2026-02 Accompanies the theory paper. MIT.
UW-Madison-Lee-Lab/looped-tf Looped transformers, length generalisation
arXiv:2409.15647
14 · 2025-04 Small research codebase.

Papers with no public implementation

Benchmarks and data

There is no shared benchmark for this field

ProsQA, a synthetic DAG-shaped multi-hop task with roughly 17.9k train and 300 validation examples at 3–6 hops, is the only benchmark purpose-built for latent reasoning, and it ships as JSON inside the Coconut repository rather than as an independent dataset. Everything else is borrowed: ProntoQA (synthetic first-order-logic QA, predating the field), GSM8K-Aug (385k augmented GSM8K samples with intermediate expressions stripped, now the de facto training corpus for implicit CoT), GSM-Hard, and the standard suites — MATH-500, AIME, SVAMP, BIG-Bench Hard, CommonsenseQA, ARC-C, GPQA.

None of these was designed to isolate whether a model is genuinely computing latently rather than taking a shortcut. Every critique paper in this area has had to build its own diagnostic. Cross-paper comparison is correspondingly unreliable, and this is a real, actionable gap rather than a search failure.

Framework support

None. No evidence that Transformers, vLLM, TGI, SGLang or llama.cpp exposes continuous-thought decoding, recurrent-depth loop control, or pause-token mechanisms as a documented first-class API, the way they expose KV-cache quantisation or speculative decoding. Huginn and Ouro ship custom modeling files loaded through trust_remote_code=True, which is a packaging convenience rather than a library feature. The absence is itself informative: every method here remains a bespoke one-off.

Curated lists

Working on something that belongs here? Open an issue on the repository.