AI Coding Live News

Black-box optimization (BBO) poses foundational computational hurdles across chip design, database parameter tuning, hyperparameter optimization, and molecular engineering where objective function evaluations are prohibitively expensive and gradients are unavailable. Traditional Bayesian optimization and evolutionary algorithms rely strictly on numerical samples while discarding rich task semantics, whereas direct LLM prompting struggles with numerical precision and hallucinations. Researchers from Nanjing University's National Key Laboratory for Novel Software Technology (LAMDA Group) introduce AgenticBBO-Bench (arXiv:2610.12183), the first unified cross-domain benchmark for evaluating LLM agents in black-box optimization. Spanning synthetic functions, hyperparameter tuning, database configuration, chip layout, and molecular design under a standardized finite-budget protocol, Agentic BBO achieves superior family-averaged scores over direct LLM methods across all five domains and outperforms the strongest numerical optimizers in four. Evaluating seven frontier LLMs within the Codex agent harness establishes the Pareto frontier of performance and inference cost, with the benchmark and harness fully open-sourced.

Modern AI agent systems and autonomous frameworks rely heavily on greedy search or reinforcement learning for self-evolution (such as harness synthesis, workflow routing, and prompt optimization). However, in non-convex and deceptive optimization landscapes, greedy pruning prematurely discards mutations that temporarily underperform, stranding systems in local optima. Researchers from Ant Group Research, Peking University, and Beijing Academy of Artificial Intelligence (BAAI) unveil Mara Chain (arXiv:2609.35855), an evolutionary optimization framework that rethinks failures as essential stepping stones. Mara Chain operates via three interconnected mechanisms: a stepping-stone archive combining semantic embeddings and execution traces for novelty preservation, a hierarchical mutator driven by backward failure attribution, and a Pareto-optimal non-dominated sorting selection module. On the AppWorld benchmark, Mara Chain achieves up to a 20.5% performance improvement over leading baselines (GEPA, ACE, SkillOpt-Lite) while requiring 65.5% fewer rollouts. On TerminalBench 2.1, it outstrips AHE and Meta-Harness by 20.2 and 22.5 percentage points respectively. The team has open-sourced the implementation under the AntOmniEvo repository.

CloudflareDev announced Union Alpha as The Unbiased Co.'s Pareto model, claiming intelligence comparable to Fable and Astra at one-quarter the cost. Bringing that cost-performance point into Cloudflare AI gives coding agents and other high-frequency workflows another model-routing option.

OpenRouter confirmed that the stealth model previously listed as Union Alpha is TheUnbiasedCo Pareto. Community detective work and demand shortened the stealth period; the model remains available on OpenRouter at openrouter.ai/unbiased/pareto.

Arena.ai (@arena) reports Agent Arena results: DeepSeek-V4.1-Flash (Max) posts about +4.87% net improvement at roughly $0.07 median task cost — lowest among the top-3 open models — retaining ~98% of Hy4 preview’s net gain at ~73% lower cost and ~76% of Kimi K3 (Max) performance at ~92% lower cost.

Databricks shipped DeepSeek 4.1 Flash on Unity Gateway the day it launched. On OfficeQA Pro V2 it sits on the Pareto frontier, so teams can point governed Lakehouse data at the model for apps and agents under one access, spend, and performance layer.

Cognition introduced SWE-2, calling it their closest model yet to the frontier: on leading evals it matches recent frontier models at up to ~70% lower cost. They say they scaled RL to multiple trillions of parameters with a refined recipe that pushes the Pareto curve on both capability and cost. SWE-2 already powers products such as Devin Voice.

LMArena: OpenAI GPT-6 Astra (Max) takes #1 on Code Arena WebDev at 1797 pts, about +35 over Claude Fable 5.1 (Max) at 1762 and ahead of Claude Opus 5 (Max) and GPT-5.6 Sol (xHigh). It also reshapes the quality-price Pareto frontier at ~$40/Mtoken matching latest Claude pricing, on agentic web-dev workflows with multi-step reasoning and tool use.

Artificial Analysis shipped Intelligence Index v4.2: adds private agentic knowledge-work eval AA-Briefcase and long-document GDP.pdf, doubles held-out weighting, upgrades grading, and drops saturated GPQA Diamond. Claude Fable 5.1 leads; GPT-6 Astra is second with ~4pt gain over GPT-5.6 Sol and strong output-token Pareto efficiency. A major interim benchmark refresh for model selection and cost/quality tradeoffs.

Arena.ai reports Claude Fable 5.1 (Max) takes #1 on Code Arena: WebDev at 1765 pts—up from Fable 5 (Max) at #8 / 1628 (+137 overall)—and leads Qwen3.8-Max-0902 by ~77 pts. It ranks #1 in five of six domains with large gains in Gaming, Simulations, and Reference-Based Design, and at ~$40/MToken reshapes the higher-price Pareto frontier.