AI Coding Live News

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.

@DanKornas highlighted TokenRhythm/opensquilla, a token-efficient microkernel AI agent for builders. SquillaRouter scores each turn locally on length, language, code, keywords, and semantic embeddings, then routes across four tiers to the cheapest capable model; the registry targets 20+ LLM backends with primary-plus-fallback. Fifteen bundled skills load on demand, and the agent can act as an MCP client or server. Local Markdown notes add SQLite full-text search and sqlite-vec semantic recall, with three sandbox policy tiers plus human-in-the-loop approvals for sensitive tools. Apache 2.0.

Alibaba Qwen released open-weight Qwen3.8-Flash, a multimodal Mixture-of-Experts model billed as an early preview of the Qwen4 architecture. The production-scale checkpoint is 125B parameters plus 51B N-gram embeddings, with only 6B activated per token. The new stack combines Gated DeltaNet plus Qwen Sparse Attention, Gated Residual branches, N-gram embeddings that can live in host memory, and a refined Muon optimizer. Qwen said it trained at about one-ninth the cost of Qwen3.7-Plus while beating it across the board, especially on coding and office work: 58.7 on DeepSWE 1.1, 62.5 on SWE-bench Pro, 73.9 on CoWorkBench, 84.5 on AndroidWorld, and 95.7 on MathVision with CI. Native context is 262K tokens, extendable to 1M with YaRN; at 1M context the QSA kernel is up to 7.6x faster in prefill and 4.9x faster in decode. QwenCloud lists roughly $0.15 input / $0.47 output per million tokens, and day-0 runtimes arrived from NVIDIA NeMo, SGLang, Unsloth, OpenRouter, and OpenCode Go.

Moonshot AI released a specialized Hybrid Code Search API powered by Kimi k1.5. Combining dense vector semantic embeddings with sparse AST inverted indexes, it resolves cross-file call hierarchies across 10-million-line monolithic codebases with search latency under 80ms.

Milvus introduced StructArray in version 3.0, allowing multiple embeddings (video clips, passages, image patches) to remain attached to a single parent record, enabling native MaxSim search and ColBERT/ColPali late-interaction models.

Cursor deployed its fourth-generation Code Embeddings v4 model. Featuring compact quantized vector representations, the update cuts background indexing RAM usage by 70% while improving cross-repository symbol resolution accuracy by 40% during contextual agent codebase retrieval.

Alibaba Qwen confirmed that the 125B multimodal MoE Qwen3.8-Flash-Next can now run locally on about 75GB of RAM via Unsloth GGUFs, with N-gram embeddings offloaded to CPU or unified memory for near-VRAM speed. It is the day-0 local path for the Qwen4 architecture preview weights.

Alibaba Qwen open-sourced Qwen3.8-Flash, a multimodal MoE with 125B parameters plus 51B n-gram embeddings and only 6B active per token. It scores 58.7 on DeepSWE 1.1 and 62.5 on SWE-bench Pro, was trained at 1/9 the cost of Qwen3.7-Plus, and ships Qwen3.8-Flash-Next as an early look at the Qwen4 architecture.

Cursor rolled out Background Indexing 2.0 with incremental AST optimization and lightweight embeddings for 10M+ line codebases.