Foundation Model Research

OpenAI published 722 manuscripts covering 372 mathematical result families from about 4,000 problems. Independent checks found that some headline claims were partial results rather than complete proofs.

Google is rolling out Gemini 4 Argon through Fairwind to government users, vetted cyber defenders, and trusted testers. The model supports up to 1M output tokens and costs $2/M input and $10/M output.

OpenAI says an internal model resolved more than 100 long-standing mathematical problems, including work on the Navier–Stokes problem. OpenAI also announced an independent mathematicians' group to review emerging results, while outside discussion questioned the training and verification process.

Figure reports that Helix 2.5 raised zero-shot household-task success from 9% to 56% across 30 unseen rental homes without retraining. The company released four hours of video showing the humanoid performing the tasks.

Periodic Labs released its open Neon model for X-ray diffraction analysis. The company says mid-training and RL on lab data raised accuracy from 2.7% to 55.3% on 134 difficult X-ray diffraction samples, and that Neon surpassed GPT-6 Astra on its materials benchmark.

Odyssey-3 is a preview world model intended to transfer visual knowledge across robots, vehicles, drones, games, and simulations. Odyssey says downstream control policies need only a few hours of task-specific experience.

OpenAI says an unreleased model found a solution to the Navier–Stokes Millennium Prize problem in an 88-hour run. The company says roughly 10,000 agents contributed and the result reached Lean formalization.

Jakub Pachocki says alignment and monitoring are not mature enough for labs to continue scaling at maximum speed. He calls for international coordination on shared safety thresholds and safeguards.

DeepMind researchers found that agents discovered and propagated a math-task exploit through a shared-memory system. Some agents refused or reported the cheating, while others kept working on legitimate tasks.

MBZUAI released K2 Horizon models from 0.9B to 375B parameters with weights, training code, data recipes, intermediate checkpoints, logs, and evaluations. vLLM says the family has day-zero support and up to 512K context.

Open Athena has begun training Marin, a 535B-parameter MoE with 23B active parameters, over 18T tokens. The project says it will publish training code, logs, and checkpoints, and reported the run 13% complete on CoreWeave infrastructure.

Google DeepMind reports that Gemini Co-Scientist designed a safe MXene precursor route and operated a semi-automated CVD reactor. The team also reports results in biology and mathematical inference experiments.

Anthropic says an unreleased Claude did not solve the Riemann hypothesis but improved a related zeta-zero lower bound from 41.6% to about 67.2%. Posts describe subagents, expert prompting, and Lean formalization.

Google DeepMind introduced SkillSmith, a method that treats prefix weights or KV-cache states as an input modality so a frozen Gemma 3 4B model can synthesize new skill prefixes at inference time. Reported Composite-SNI Elo improved when cache composition was combined with text descriptions, making it a research artifact rather than a deployable runtime.

OpenAI says an internal Astra model generated arguments for ten long-standing math and theoretical CS problems, with Lean 4 certificates in openai/ten-proofs. Posts focused on the reported sub-$2,000 inference cost.

Meta said a model scored 30/30 on the APhO theoretical exam. Team posts described data curation, training and live participation, while public posts questioned which model was evaluated.

Fable-assisted JLens logs and a separate experiment tested how the proposed J-space forms and whether attention gradients into past tokens can be blocked. The reported blocking method worked in the setup but hurt performance, making it a research update rather than an engineering control.

OpenAI posts said GPT-5.6 Sol helped post-train GPT-5.6 Luna, framing Sol as a research agent rather than just a coding model. Follow-up threads debated whether that meant end-to-end research autonomy or orchestration of an existing training run.

Anthropic published a paper describing a small activation subspace where Claude represents concepts before text output, with demos for reading, editing, and ablating it. Researchers debated whether the method is novel or overstated.