PapersWithCode

Trending AI research papers with code, datasets and leaderboards.

[PapersWithCode @paperswithcodee]

https://t.co/uaLimd7npA

[PapersWithCode @paperswithcodee]

BizBench: Unveiling Hidden Challenges in Deploying Long-Horizon Agents for Real-World Business

BizBench evaluates long-horizon agents on 50 realistic payments tasks with 789 policy documents and 601 tools, checking intermediate validity alongside final outcomes.

[PapersWithCode @paperswithcodee]

https://t.co/A7YkTgrln4

[PapersWithCode @paperswithcodee]

Terminal Bench 2.1 is the fastest-growing benchmark on Papers with Code.

58 papers reported results on it in the last 90 days, up from 21 in the 180 days before.

It updates 26 tasks for more reliable agent evaluation.

[PapersWithCode @paperswithcodee]

https://t.co/H9DRR47CTi

[PapersWithCode @paperswithcodee]

DecisionPivot: A Diagnostic Benchmark for Decision Models in LLM Systems

DecisionPivot tests whether decision models handle controlled variations, rank confidence correctly, and calibrate selective predictions—not just answer individual items.

[PapersWithCode @paperswithcodee]

https://t.co/kM1FQk8V5x

[PapersWithCode @paperswithcodee]

Learning the 4D World through Point Trajectory Forecasting

VPM jointly forecasts multi-view videos and persistent 3D point trajectories, anchoring appearance changes across cameras to shared scene motion.

[PapersWithCode @paperswithcodee]

@Apple https://t.co/eRCPgzBJv7

[PapersWithCode @paperswithcodee]

Congrats @Apple on achieving SOTA on RealWorldQA!

GPT-6 Astra + OpenVLHarness scores 92.16 Accuracy.

That puts it ahead of Qwen3.8-Flash-Next at 88.5.

[PapersWithCode @paperswithcodee]

https://t.co/f8INDlf4V7

[PapersWithCode @paperswithcodee]

DiscoveryRSI: Recursively Learning to Discover Valuable Hypotheses

DiscoveryRSI recursively improves scientific hypothesis generation through multi-agent evolutionary search, credit assignment, workflow replay, and rediscovery.

[PapersWithCode @paperswithcodee]

https://t.co/e1Yb2OVsOq

[PapersWithCode @paperswithcodee]

Urbex: Agentic Spatial Grounding for City-Scale 3D Scenes

Urbex casts city-scale 3D grounding as active spatial search rather than candidate ranking.

[PapersWithCode @paperswithcodee]

https://t.co/KEMwfiqPcO

[PapersWithCode @paperswithcodee]

UltraBERT: Beyond the Limit of BERT Pre-Training

UltraBERT revisits encoder pretraining, finding that extra compute saturates after a few trillion tokens and that masking ratios should differ by task.

[PapersWithCode @paperswithcodee]

https://t.co/onPe3Fyn7w

[PapersWithCode @paperswithcodee]

StreamEval: Evaluating Interactive Streaming Video Generation Models

StreamEval benchmarks interactive streaming video across five tasks, measuring execution, state consistency, visual fidelity, and audiovisual coordination.

[PapersWithCode @paperswithcodee]

https://t.co/jJEKBjOcUT

[PapersWithCode @paperswithcodee]

Deep Research Agents is the fastest-rising task on Papers with Code.

43 papers in the last 30 days, 4.2x its share of new papers in the 180 days before.