cd /news/artificial-intelligence/energy-based-reasoners-385x-more-eff… Β· home β€Ί topics β€Ί artificial-intelligence β€Ί article
[ARTICLE Β· art-171692] src=energy-based-reasoners.github.io β†— pub= topic=artificial-intelligence verified=true sentiment=↑ positive

Energy-Based Reasoners: 385x more efficient inference

Researchers introduced Energy-Based Reasoners (EBRs), reasoning models that learn an energy function over latent reasoning states and treat looped updates as gradient descent on that energy, reaching 99.9% accuracy on Sudoku-Extreme with 1.8M parameters while using 345Γ— less inference compute than the next-best model. The EBR energy scores candidate predictions without an external verifier, its gradient guides refinement and supplies a stopping signal, and exploration-based training shapes the energy landscape. On N-Queens (8Γ—8 and 10Γ—10) and Graph Coloring (8- and 10-vertex), EBRs reported state-of-the-art solution coverage with near-100% single-sample accuracy, including 0 conflicts on both Graph Coloring sizes.

recommendation

Replace fixed loop counts with an energy-gradient stopping threshold so easy inputs exit after a few descent steps and hard ones keep refining; the source reports 345Γ— lower inference compute at 99.9% Sudoku-Extreme accuracy.

more in playbook: Inference and self-hosting
read4 min views10 publishedNov 18, 2026
Energy-Based Reasoners: 385x more efficient inference
Image: source

TL;DR – We introduce Energy-Based Reasoners (EBRs), reasoning models that learn an energy function over latent reasoning states and formulate looped updates as gradient descent on this energy. Its value scores candidate predictions without requiring an external verifier, while its gradient guides refinement and provides a natural signal for stopping. With only 1.8M parameters, EBRs achieve 99.9% accuracy on Sudoku-Extreme while using 345Γ— less inference compute than the next-best model. On N-Queens and Graph Coloring, EBRs achieve state-of-the-art coverage while maintaining near-100% single-sample accuracy.

Looping is a powerful inductive bias for reasoning. By repeatedly iterating a recurrent block, looped models can scale their effective depth and enable multi-step reasoning in latent space. However, existing looped models are not inherently guided by an explicit measure of progress, making it difficult to determine when to stop looping and how to improve upon the current state. To enable this, we introduce Energy-Based Reasoners (EBRs), which learn an energy function over latent reasoning states and formulate looped updates as gradient descent on this energy. The energy scores latent states, while its gradient guides refinement and provides a signal for stopping. To train EBRs effectively, we further introduce exploration-based training to discover additional modes and to better shape the energy landscape over a broader set of states. Across four algorithmic reasoning benchmarks, EBRs match or outperform prior recurrent reasoning models, achieving 99.9% accuracy on Sudoku-Extreme while using 345Γ— less inference compute than the next-best model and maintaining comparable training compute. On all multi-solution benchmarks, EBRs achieve state-of-the-art coverage of distinct valid solutions while maintaining near-100% single-sample accuracy. These results suggest that energy-based models are a promising approach for strong reasoning performance.

Existing looped reasoners update their latent state without any explicit assessment of it. An EBR learns an energy function that guides refinement, stopping and selection.

Score. The energy rates the compatibility between the reasoning state and the initial board . Lower means better compatibility.

Refine. The gradient turns the score into an improvement direction.

Stop. The state becomes stationary once the gradient is small, so the gradient acts as a stopping criterion.

Select. Among multiple candidate solutions, we pick the one with the lowest energy.

Running more descent steps refines each answer further, and running more trajectories gives the energy more candidates to choose the lowest-energy one from.

Each trajectory stops once its energy gradient falls below a threshold, so easy puzzles finish in a few steps and hard ones keep going.

Method # Params Sudoku-Extreme ↑ Maze-Hard ↑
Single sample (%)
TRM 5–7M 87.4 85.3
EqR 5–7M 93.0 –
FPRM 7M 94.2 87.0
GRAM 10M 86.6 –
Looped flows 5–7M 97.9 86.7
EBR Ours 1.8M 99.4 87.4
Multiple samples (%), number of samples N in parentheses
PTRM 5–7M 98.8(100) 86.7(100)
EqR 5–7M 99.8(128) –
GRAM 10M 96.1(20) –
Looped flows 5–7M 99.3(5) 86.9(5)
EBR Ours 1.8M 99.9 (5) 88.3 (5)

EBR acts like stochastic gradient descent and can capture many different solutions. Coverage captures the share of each input's distinct valid solutions that 20 samples find.

Method N-Queens Graph Coloring
8Γ—8 10Γ—10 8-vertex 10-vertex
Single sample Pass@1 (%) ↑ Conflicts ↓
TRM 66.8 17.5 109.3 170.7
GRAM 99.5 89.7 8.3 31.3
EBR Ours 100 99.5 0 0
20 samples Cov@20 (%) ↑ Cov@20 (%) ↑
TRM 22.6 2.0 22.3 6.8
GRAM 89.7 57.5 85.8 51.3
EBR Ours 98.1 92.0 88.6 54.5
@misc{chen2026energybasedreasoners,
  title         = {Energy-Based Reasoners: Your Recurrent Reasoner Should Be an Energy-Based Model},
  author        = {Chen, William Xingxu and Gladstone, Alexi and Li, Zeyu Michael and Ji, Heng and Du, Yilun},
  year          = {2026},
  eprint        = {2611.08977},
  archivePrefix = {arXiv},
  primaryClass  = {cs.LG},
  url           = {https://arxiv.org/abs/2611.08977}
}
── more in #artificial-intelligence 4 stories Β· sorted by recency
── more on @energy-based reasoners 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain β€” perfect for shipping the agent you just read about.

$git push zahid main
β†’ Live at https://your-agent.zahid.host βœ“
Get free account β†’ Pricing
from €0/mo Β· no card required
LIVE [news/energy-based-reasone…] indexed:0 read:4min 2026-11-18 Β· β€”