AICoder

AI Coding Live News

First-hand AI coding updates from 𝕏/Twitter frontier labs and researchers, distilled by AI with entity benchmarks.

Explore live on 𝕏 ↗

AI Coding Live News

@
@huggingface@huggingface·
👀100
🔥94

Judge-Free Test-Time Scaling Tops WebArena Infinity at 74.6%: Salesforce Introduces CLIFT, Alongside MMSearch-OPD and SCOPED-8B Beating 32B VLMs

Three new frameworks advance web and multimodal agent post-training: Salesforce AI Research's CLIFT (arXiv:2610.06829) certifies URL-conditional self-verification questions for dense RL rewards and judge-free Conformal Trajectory Selection (CTS), propelling Gemma-4 31B to a 74.6% SOTA on WebArena Infinity (beating Gemini 3 Flash at 70.1%) and lifting GPT-5.5 to 53.7% on VisualWebArena. Meanwhile, MMSearch-OPD (arXiv:2610.22240) boosts 9B multimodal deep search by +15.3 points to 60.2%, and Northeastern University's SCOPED-8B (arXiv:2610.22217) uses skill-decoupled on-policy expert distillation to reach 73.13% across 8 benchmarks—surpassing Qwen3-VL-32B (72.01%) and Claude Opus 4.6 (72.86%).

Judge-Free Test-Time Scaling Tops WebArena Infinity at 74.6%: Salesforce Introduces CLIFT, Alongside MMSearch-OPD and SCOPED-8B Beating 32B VLMs

Key Takeaways

  • Salesforce's CLIFT conformal self-verification propels Gemma-4 31B to 74.6% on WebArena Infinity (+12.8 pts over base, beating Gemini 3 Flash at 70.1%) with judge-free test-time trajectory selection
  • Transferring the open-trained CLIFT question bank to GPT-5.5 sets a 53.7% SOTA on VisualWebArena, while MMSearch-OPD boosts 9B multimodal deep search accuracy by +15.3 points to 60.2%
  • Northeastern University's SCOPED-8B reaches 73.13% average accuracy across 8 VLM benchmarks via skill-decoupled on-policy expert distillation, beating Qwen3-VL-32B (72.01%) and Claude-Opus-4.6 (72.86%)
Read full story
Linked Models & Agents:
#CLIFT#WebArena#VisualWebArena#Salesforce-AI#MMSearch-OPD#SCOPED
H
Hugging Face Daily Papers@HuggingFace·
👀58
🔥99

GitSwarm Unveils Decentralized Compounding Inference: Multi-Agent Collaboration via Git Branching and Atomic Commits

Tackling long-horizon problem solving and scientific discovery requires computation to compound across successive attempts. While humans build upon partial proofs, failed experiments, and prior hypotheses, current test-time compute paradigms (e.g., MCTS and Best-of-N) isolate search trajectories into independent silos, discarding intermediate reasoning artifacts upon branch failure. Researchers introduce compounding inference and its open-source instantiation GitSwarm (arXiv:2610.04862). GitSwarm establishes an asynchronous multi-agent coordination architecture where homogeneous LLM agents collaborate through structured persistent memory hosted in a shared, branchable Git repository. Intermediate artifacts are preserved as atomic commits, while explicit semantic dependencies link cross-branch contributions, enabling agents to inspect, branch, verify, and synthesize past discoveries. Evaluated on rigorous long-horizon benchmarks, GitSwarm paired with GPT-5.5 solves all 30 problems on IMOProofBench-Advanced in a single execution run, and scores 79.4% on ProgramBench compared to 65.1% for the strongest reported baseline under identical budgets.

Key Takeaways

  • Introduces Compounding Inference and GitSwarm, converting transient test-time compute into persistent knowledge assets
  • Employs Git repositories as multi-agent memory, tracking atomic commits, semantic dependencies, and cross-branch merges
  • Solves 100% of IMOProofBench-Advanced problems with GPT-5.5 in a single run and achieves 79.4% on ProgramBench
Read full story
Linked Models & Agents:
⚡ GitSwarm⚡ GPT-5.5⚡ Compounding Inference🤖 Swarm-Worker🤖 Git-Reviewer🌐 Open Source Research🌐 Hugging Face
#GitSwarm#Compounding Inference#Coding Agents#Multi-Agent#Test-Time Scaling#Open Source
K
KAIST AI & Research@KAISTPR·
👀75
🔥95

KAIST Unveils FlyBy: Overcoming Blind Thinking in Small Reasoning Models via Cost-Aware RL, Outperforming 14B at 63% Lower Cost

While test-time compute scaling has propelled reasoning capabilities in large models, small reasoning models (sRMs, 4B-8B) frequently fail when scaling thinking tokens: self-refinement merely consolidates probability onto already reachable hypotheses, failing completely when missing parametric knowledge. KAIST researchers introduce FlyBy, a selective querying framework that formally decouples execution bottlenecks (recoverable through internal reflection) from knowledge bottlenecks (requiring external teacher queries). Optimized via cost-aware reinforcement learning, FlyBy teaches small models when to think, when to ask, and how much compute budget to expend. Evaluated across 1,158 hard problems on six benchmarks, FlyBy-4B reaches 45.96% pass@8, outperforming the much larger Qwen3-14B (41.64%) while slashing serving costs by 2.7x (a 63% reduction).

KAIST Unveils FlyBy: Overcoming Blind Thinking in Small Reasoning Models via Cost-Aware RL, Outperforming 14B at 63% Lower Cost

Key Takeaways

  • Diagnosing Blind Thinking in sRMs: Empirical interventions demonstrate that self-refinement primarily concentrates probability mass on pre-reachable states; lacking key facts, extra test-time compute leads to hallucination loops.
  • Execution vs. Knowledge Bottleneck Decoupling: FlyBy introduces formal state interventions to separate execution errors (amenable to reflection) from knowledge bottlenecks (requiring external model invocation).
  • 4B Surpasses 14B at 63% Lower Cost: On 1,158 difficult multi-domain problems, FlyBy-4B scores 45.96% pass@8 (beating Qwen3-14B's 41.64%) at 2.7x lower serving cost, while FlyBy-8B scales to 51.81%.
Read full story
Linked Models & Agents:
⚡ FlyBy-4B⚡ FlyBy-8B⚡ Qwen3-14B⚡ Qwen3-8B⚡ DeepSeek-R1 (Open Reasoning)🤖 Small Reasoning Model🤖 Selective Query Agent🤖 Hybrid Edge-Cloud Agent🌐 KAIST🌐 Hugging Face🌐 arXiv
#Reasoning#Test-Time Compute#Small LLMs#Reinforcement Learning#Edge AI#CoT
H
Hugging Face Daily Papers@HuggingFace·
👀88
🔥96

Tsinghua Open-Sources TaH2: Adaptive Looped Transformers Boost Test-Time Scaling Slope by 53% on AIME

While looped transformers achieve high parameter efficiency by iteratively recycling layers for latent compute, fixed-depth looping expends redundant iterations on simple tokens and suffers from performance saturation. Tsinghua University's NICS lab developed TaH2, an adaptive looped framework that co-trains the backbone and an iteration decider via lookahead depth supervision. Evaluated on the rigorous AIME competition benchmarks, TaH2 accelerates the accuracy-compute scaling slope by 53% (2.74 vs 1.79) and surpasses non-looped baselines by +3.4 to +3.9 points at matched test-time compute.

Tsinghua Open-Sources TaH2: Adaptive Looped Transformers Boost Test-Time Scaling Slope by 53% on AIME

Key Takeaways

  • Adaptive Looping via Lookahead Depth Supervision: Discards rigid, uniform layer looping by training a lightweight iteration decider using online labels that predict whether subsequent recurrent passes genuinely improve token predictions.
  • 53% Steeper Test-Time Compute Scaling Slope: On challenging AIME competition benchmarks, improves the accuracy-to-compute slope from 1.79 to 2.74, outpacing non-looped baselines by +3.4 points at matched decoding FLOPs.
  • Overcoming Looping Saturation: While existing recurrent transformers plateau beyond shallow depths, TaH2's performance margin over dense models expands continuously from +2.8 points at depth 2 to +3.9 points at depth 8.
  • Complete Codebase Open-Sourced: Full post-training recipes, decider checkpoints, and inference evaluation harnesses released at github.com/thu-nics/TaH.
Read full story
Linked Models & Agents:
#Looped Transformer#Test-Time Scaling#AIME#Mathematical Reasoning#Open Source
ADSponsored
You've caught up with all stories