42
Smart search ranks by meaning as well as keywords (one row per story, last 45 days).
Meet Redis LangCache: A Managed Semantic Cache That Cuts LLM API Costs by Up to 90% and Returns Cache Hits Up to 15x Faster
Redis launches LangCache, a managed semantic cache matching LLM prompts by meaning, cutting API costs up to 90% and returning hits up to 15x faster.
Redis LangCache is a fully managed semantic caching service in public preview on Redis Cloud, accessed via a REST API with Python and JavaScript SDKs. It embeds incoming prompts, vector-searches stored entries, and returns a cached response when similarity clears a configured threshold, skipping the LLM call entirely. Redis claims up to 90% cost savings and up to 15x faster cache hits; a demo run showed 0.37 seconds versus 2.232 seconds direct inference (about 6x) with zero LLM tokens. Customer Mangoes.ai reports a 70% hit rate, 70% lower LLM spend, and 4x faster responses on a patient-care voice app.
30
Tracking Shai-Hulud: Inside the ChainDrop NPM Worm
ChainDrop, a Mini Shai-Hulud variant tied to TeamPCP, worms through npm via a compromised maintainer account and anchors C2 in an Ethereum smart contract.
On August 4, 2026, the self-propagating ChainDrop worm entered the npm ecosystem through a compromised maintainer account, poisoning packages including keyv, cacheable, and flat-cache. ChainDrop is a variant of Mini Shai-Hulud linked to the actor TeamPCP. It anchors its C2 infrastructure in an Ethereum smart contract, allowing domain rotation with a single blockchain transaction and defeating domain-based blocklists. The compromised keyv maintainer's GitHub account caused the project's own GitHub Actions release pipeline to publish malicious package versions.
68
A Three-Layer Caching Architecture for Low-Latency LLM Web Search on Commodity CPU Hardware
OreoLook's three-layer Redis caching architecture cuts redundant LLM calls and embedding work for CPU-hosted web-search answer generation.
The paper describes a three-layer caching architecture for OreoLook (formerly lixSearch), an open-source LLM answer engine: a Redis session context window with Huffman-compressed disk overflow, a semantic query cache matching rephrasings via embedding cosine similarity, and a URL embedding cache deduplicating embedding computations. Deployed on a single 8-vCPU Intel Cascade Lake server with 30 Hypercorn workers across three containerized replicas, it achieved an 89.3% aggregate Redis keyspace hit rate, 0.1 ms read latency, and 1.38 MB memory overhead. An LRU eviction daemon migrates idle sessions to disk and rehydrates them for resumption hours or days later.
34
Alleged TeamPCP Hackers Charged in Australia Over Major Supply Chain Attacks
Australian Federal Police charges two men with 14 offences over TeamPCP supply chain backdoors in Trivy, Checkmarx KICS, and LiteLLM affecting 1,000+ organizations.
The Australian Federal Police charged two Western Australian men, aged 21 and 23, with 14 offences for their alleged principal roles in TeamPCP's March 2026 supply chain attacks; they appeared in Perth Magistrates Court on August 27, 2026. The group stole publishing credentials from trusted open-source projects and pushed poisoned releases across five ecosystems - GitHub Actions, Docker Hub, npm, PyPI, and OpenVSX - with LiteLLM's unpinned Trivy install enabling token theft and backdoored LiteLLM releases. The FBI said the malicious code potentially compromised more than 1,000 organizations, enabled theft of over 500,000 credentials, and exfiltrated at least 300 GB of data. Oligo Security linked the group's infrastructure to activity back to 2020, previously tracked as TA-NATALSTATUS and IronErn.
72
Grouped Value Attention: Efficient KV Caching via On-Demand Key Reconstruction
Grouped Value Attention stores grouped values and reconstructs content keys via a learned linear map, cutting KV-cache size about 45-47% versus GQA.
GVA stores only grouped values and reconstructs content keys with a learned linear map absorbed into the query at decode time, while a small shared decoupled RoPE channel preserves positional information via a separately cached positional key. At 350M parameters trained on 30B FineWeb-Edu tokens, the 16-dimensional positional variant scores 44.18 average accuracy across five tasks versus 44.36 for GQA and 43.88 for MLA. Custom decoding kernels are in development with an open-source release planned.
42
60
Does Your Agent's Memory Survive a Model Upgrade? A Controlled Study of Memory Portability
A controlled study finds agent memory portability varies sharply: fixed-schema knowledge graphs survive model swaps while compressed notes degrade.
The study compares preserving an agent's history as raw long context, RAG chunks, compressed natural-language notes, or fixed-schema knowledge graphs across model upgrades, using 48 synthetic histories and two open-weight sub-10B-parameter models. Fixed-schema KG accuracy changed by only +0.0004 ± 0.0020 after a writer swap, while compressed NOTES shifted asymmetrically by +9.91 or -13.28 percentage points depending on migration direction. Mixed 50/50 embedding migrations captured only 4.96 of an 11.90-point RAG re-embedding gain; 80% of the NOTES deficit came from information lost at construction, and 81% of the RAG deficit from retrieval failures. Store-only repair of NOTES failed to reach 90% recovery in all 48 cases, while retaining raw histories enabled recovery in 34 of 48 for one direction.
32
Mozilla Revokes Firefox and Thunderbird Linux Signing Key After Key Lands in Private Repo
Mozilla revoked the Firefox and Thunderbird Linux RPM signing subkey after an unencrypted copy landed in a private repo; no unauthorized access found.
Mozilla revoked the OpenPGP signing subkey (fingerprint 09BE ED63 F346 2A2D FFAB 3B87 5ECB 6497 C1A2 0256) used for Firefox and Thunderbird Linux downloads after an unencrypted copy was committed to one of its private repositories, citing reason code 2, 'key material has been compromised.' Audit records showed no sign of unauthorized access, and a replacement subkey (827E 6586 0867 9618 CD34 9F93 678E 455D 7676 7AA3) valid until August 5, 2028 was published. Users who verify signatures manually or install from Mozilla RPM packages may need to import the new key and remove the old one. The rotation came roughly seven months ahead of Mozilla's usual two-year subkey cycle.
45
BeaconKV: Key-Value Cache Compression Guided by Beacon Queries for Efficient Large Reasoning Model Inference
BeaconKV introduces training-free KV cache compression using beacon queries, cutting long-reasoning inference memory up to 5.8x while preserving accuracy.
The paper shows recency-based KV cache compression assumptions fail in long-horizon reasoning because Thought Revisiting Tokens (TRT) re-attend to distant context such as early task-solving plans. TRT queries cluster into a small number of similarity groups, which BeaconKV exploits by maintaining compact beacon query representatives to anticipate revisited KV pairs without storing full query history. The training-free method achieves up to 5.8x memory reduction and over 4.3x throughput improvement across four open-source large reasoning models while nearly preserving full cache accuracy.
33
30
Beyond the Turing threshold: Productive grammars generate essentially undecidable languages
A theoretical paper designs formal grammars that emulate Post's productive sets, generating languages that are provably beyond Turing decidability.
The paper elaborates on Emil Post's productive sets, which are not even semi-computable, and builds formal grammars that emulate their construction over natural numbers. The resulting languages are shown to be essentially undecidable, placing them beyond Turing decidability. This is pure computability and formal language theory with limited direct security relevance.
10
Agent Harness vs Agent Framework vs MCP: Which Layer Owns the Loop, State, Tools, Permissions, and Recovery
Architecture explainer separates agent harnesses, frameworks, and MCP by which layer owns the loop, state, permissions, and recovery.
The article distinguishes agent harnesses (OpenAI Codex, Claude Agent SDK), which own the execution loop, sandbox, permission model, and recovery; frameworks (LangGraph, OpenAI Agents SDK, Microsoft Agent Framework), which supply composable primitives; and MCP, a stateless JSON-RPC wire protocol governed by the Linux Foundation's Agentic AI Foundation since December 2025. An ownership matrix maps the execution loop, state, tool transport, permissions, recovery, sandboxing, and multi-agent orchestration to each layer. The 2026-07-28 MCP specification made the protocol fully stateless, retiring the initialize handshake and session headers.
35
From Infostealer Log to Marketplace Listing: A Technical Walkthrough of the Credential Theft Pipeline
Cyble walkthrough maps how infostealer logs move from endpoint infection through aggregation and enrichment to dark web credential marketplace sales.
Cyble breaks the credential theft pipeline into stages: infostealer execution harvesting browser credential stores, cookies, session tokens, crypto wallets, and FTP configurations; aggregation of stealer logs via C2 panels into bundled archives; parsing and enrichment against previously leaked datasets; and final listing on dark web marketplaces. Enrichment adds employer and role context that raises prices and enables credential stuffing across reused passwords. The report advises SOC teams to monitor stealer logs and marketplace chatter early rather than waiting for breach alerts.
30
PACE: Perceived-Latency-Aware Cascading Service Routing and Filler Control for QoE-Efficient Retrieval-Augmented Dialogue Serving
PACE cuts perceived latency in retrieval-augmented dialogue serving via cascading routing and filler control, reaching 0.41s P95 under load.
PACE is a serving framework for retrieval-augmented dialogue that optimizes Perceived Time-to-First-Response (PTFR) under quality and cost constraints. It combines a load-adaptive cascading router, a joint path-filler controller, and volatility-aware cache admission, deployed on a humanoid-robot sales service. On 75k CarQA requests, the cascade halved pure-LLM P95 PTFR (0.29s vs 0.53s) and the adaptive controller reached 0.41s P95, 2.4x better than RAG at high load; filler calls dropped 94% and stale answers fell from 86% to 0%.
20
Rare Not Random Using Token Efficiency for Secrets Scanning
Researcher proposes token efficiency (string length divided by BPE token count) as a better post-regex filter than entropy for secrets scanning, validated on CredData.
The post explores whether Byte-Pair Encoding tokenization can replace Shannon entropy as the primary filter for candidate secrets captured by regex in tools like Gitleaks. It defines 'token efficiency' as string length divided by token count under the cl100k_base tokenizer; secret-like strings such as GitHub tokens tokenize into many small tokens and score low, while natural text scores high. Evaluating labeled secrets from the CredData dataset shows a usable separation, with roughly 2.5 suggested as a minimum cutoff versus Gitleaks' 3.5 entropy threshold. The technique is positioned as a post-regex filtering step rather than a standalone detector.
45
Atlas: Efficient Verifiable Semantic Search
Atlas delivers zero-knowledge proofs for HNSW semantic search, verifying RAG retrieval in under a second on SIFT1M and 2.0 seconds at 100M vectors.
Atlas lets a search provider prove that a query was answered correctly against a committed HNSW index without revealing the index, addressing provider deviations like truncation or bias. It combines offline preprocessing, a fixed-size-state restructuring of HNSW with a correctness proof, and timestep-tagged batching of per-step arguments. The system proves queries in under a second on SIFT1M and 2.0 seconds at 100 million vectors while preserving plaintext HNSW recall, and proven retrieval maintains end-to-end RAG answer quality at lower cost than prior verifiable retrieval systems.
40
MicroHasTEE: Bare-Metal Haskell for Type-Level Peripheral Ownership on Armv8-M
MicroHasTEE expresses TrustZone secure and non-secure firmware as one typed Haskell program, catching peripheral-ownership inconsistencies at compile time.
MicroHasTEE is a multiparty Haskell framework that models both TrustZone firmware images as participants in a single typed program, using type-level capability ledgers to track peripheral acquisition, configuration, transfer, and finalization. MicroHs compiles the shared program twice to produce separate bare-metal Secure and Non-secure firmware images. The framework rejects inconsistent resource use, post-configuration attribution changes, wrong-domain callbacks, and calls to unregistered Secure services. A door-lock case study on an STM32U5 Nucleo board produced images of 232.7 KiB and 228.4 KiB of flash with roughly 220 KiB of SRAM per domain.
30
Private Information Retrieval With Arbitrary Privacy Requirements: Introduction and Capacity Results
Researchers formulate private information retrieval under arbitrary graph-based privacy requirements, deriving capacity bounds and introducing pyramid storage graphs.
The paper generalizes classical private information retrieval (PIR) to arbitrary privacy requirements over graph-based storage systems, where each message is retrieved privately from a pre-specified server subset. The authors derive lower and upper capacity bounds for general graphs and exact capacity results for path and cyclic storage graphs. They also introduce a new pyramid storage graph structure that models symmetric message storage and replication patterns.
15
Hoss: Fast Oblivious Semantic Search with Heterogeneous GPU-CPU-TEE Architecture
Hoss uses heterogeneous GPU-CPU TEEs for oblivious semantic search, achieving up to 67x speedup over the Compass ORAM-based system.
Hoss is a first-of-its-kind oblivious semantic search system built on a heterogeneous CPU-GPU TEE architecture, hosting hot-path HNSW graph traversal in large GPU TEE private memory and offloading lower graph layers to CPU TEEs. It invokes oblivious primitives only for lower-layer accesses and adds host-access ORAM and data-dependent optimizations. Benchmarked against prior state-of-the-art Compass, the prototype achieves up to 67x speedup while maintaining high recall, with larger gains at scale.
25
VDiff-Bench: A Challenging Benchmark for Fine-Grained Image Difference Identification
VDiff-Bench, a 1,756-question benchmark, shows multimodal LLMs struggle with fine-grained image-difference identification, scoring as low as 8.7% on low-level changes.
VDiff-Bench is a multiple-choice benchmark of 1,756 four-way questions over image pairs covering 10 change categories including position, motion, color, texture, OCR/text and illumination, with curated hard negatives. Evaluation of 11 state-of-the-art open- and closed-source MLLMs shows fine-grained visual comparison remains brittle: 7-8B-scale open-source models score 52.5-70.6% on semantic changes but only 8.7-33.3% on low-level changes like noise and texture. Notably, Grok 4.3 shows a sharp performance drop on noise and texture differences, falling behind large open-source models like Kimi K2.5 and K3.
32
CertiFlash: A Formal Verification Framework for Flash Translation Layers in Computational Solid State Drives
CertiFlash provides machine-checked formal verification of SSD flash translation layers, proving isolation, integrity, and ownership invariants to prevent tenant data leaks.
CertiFlash is an open-source formal verification framework for Flash Translation Layers (FTL) in computational SSDs, mechanized in the Rocq proof assistant. It shows that a faulty FTL can corrupt device state at five surfaces (e.g., leaking data between tenants or dropping integrity tags), demonstrated on a DaisyPlus OpenSSD. Designers prove once that every operation of a general FTL model preserves a global invariant covering mapping, isolation, integrity, ownership, and allocation; new designs need only discharge five hypotheses. Across four case studies, added effort was 27-3,231 lines against a 16,489-line framework.
25
ZDI-26-587: Ashlar-Vellum Cobalt VS File Parsing Heap-based Buffer Overflow Remote Code Execution Vulnerability
ZDI advisory ZDI-26-587 details a heap-based buffer overflow RCE (CVE-2026-19781, CVSS 7.8) in Ashlar-Vellum Cobalt VS file parsing, requiring user interaction.
The Zero Day Initiative released advisory ZDI-26-587 covering a heap-based buffer overflow in Ashlar-Vellum Cobalt's VS file parsing. A remote attacker can execute arbitrary code when the target opens a malicious file or visits a crafted page. ZDI rated the vulnerability CVSS 7.8 and assigned CVE-2026-19781.
12
Molecular Déjà Vu: Digit-Level Retrieval of Published Values in Frontier Language Models
Audit of 22 frontier models finds widespread verbatim retrieval of published molecular property values, with higher reasoning increasing recall of memorized numbers.
An arXiv audit tests 22 frontier LLMs across 12 molecular regression benchmarks for verbatim retrieval of published values. More than 50% of the LLMs show verbatim retrieval on five datasets, and identical experiments are flagged 89% more often at a high reasoning level than at the lowest one. Suppressing retrieval moves model prediction errors closer together in relative terms, suggesting predictive capability is not determined solely by memorized values.
42
CHERI-D Reincarnate: efficient multicore CHERI temporal memory safety through allocation reincarnation (draft version)
CHERI-D Reincarnate adds allocation reincarnation to CHERI, quarantining generation IDs instead of memory to cut temporal memory-safety overheads on multicore systems.
The proposal extends CHERI-D temporal memory safety by quarantining exhausted generation IDs rather than memory slots, allowing freed memory immediate reuse through allocation reincarnation. It adds coherent ID caching and multicore support connecting physical coherence events to the virtually addressed ObjID buffer. The design was implemented as a hardware-software co-design across CHERI-Toooba FPGA softcore, QEMU, LLVM/Clang, and CheriBSD. Across evaluated workloads it substantially reduces memory-sweep frequency and quarantine overhead with low performance and hardware cost.
27
LLMs and Contextual Integrity
Bruce Schneier highlights two papers: the CIMemories benchmark shows frontier LLMs leak memory attributes up to 69%, and an RL method reduces inappropriate disclosures.
Bruce Schneier discusses contextual integrity in LLMs, referencing the CIMemories benchmark, which uses synthetic profiles with 100+ attributes per user to test whether models with persistent memory disclose sensitive information appropriately. Evaluation showed frontier models exhibit up to 69% attribute-level violations, with GPT-5's violation rate rising from 0.1% to 9.6% across 40 tasks and reaching 25.1% with repeated prompting, showing unstable leakage behavior. A second paper introduces a reinforcement learning framework trained on a synthetic 700-example dataset that substantially reduces inappropriate disclosure while maintaining task performance, with improvements transferring to the human-annotated PrivacyLens benchmark.
48
Explainability Assistant: A Conversational XAI Interface for Interpreting Energy Consumption Models
Researchers release Explainability Assistant, an open-source conversational XAI tool using LLM function calling, lifting intent-parsing accuracy from 76.8% to 94%.
The paper introduces the Explainability Assistant, an open-source conversational XAI system for interpreting energy consumption forecasting models such as genetic-programming symbolic regressors. It uses LLM function calling instead of rigid custom grammars, achieving 94% intent-parsing accuracy versus 76.8% for prior work TalkToModel, and adapts to different ML problem types without task-specific fine-tuning. Comparative evaluation with energy domain specialists against a traditional XAI dashboard showed improved usability, with all experts preferring the conversational interface.
25
Subgroup Packing for Batched PASTA Transciphering
Interleaved subgroup packing cuts PASTA homomorphic transciphering server cost by a median 1.60x at the expense of remaining noise capacity.
The paper studies how record word layout affects the cost of PASTA transciphering, where a server converts symmetrically encrypted records into homomorphic ciphertexts. A subgroup layout interleaves records so cyclic shifts preserve positions, reducing required displacements from 255 to 128 versus contiguous packing. In HElib across twelve paired corpora under six homomorphic keys, the median direct-to-subgroup server cost ratio is 1.60 including fresh generation, conversion, and two queries, but with less remaining noise capacity.
15
Revoked but Still Authoritative: An Empirical Study of Revocation Enforcement in Agent-Memory Systems
An empirical study finds no major agent-memory system enforces fact revocation at retrieval, causing agents to act on superseded, unsafe information.
Researchers tested five agent-memory systems across nine policy scenarios, nine models, and six defense conditions, tracking whether revoked facts are returned and acted upon. No system enforces revocation by default: revoked records are returned whenever the revocation label is visible to the retrieval layer, outrank their replacements, and lead agents to unsafe actions. The authors propose a backend-agnostic guard that sits between the agent and any memory store and withholds revoked or conflicting records at retrieval time.
38
Feature Recovery for Object Understanding After Irreversible Fire Damage
TRACE benchmark with 21.4K scenes studies post-fire object understanding; a Feature Recovery Module improves degraded-image retrieval by 12.5% and material recovery by 20.1%.
The paper introduces TRACE, a transformation-aware benchmark with 21.4K real-image-grounded synthetic scenes, 499 object identities across 189 categories, and five tasks covering degraded-object detection, pristine-state recovery, material recovery, description generation, and functional reasoning. Existing models degrade sharply: RF-DETR mAP falls 71% relative from least to most severe level, and InternVL3.5 retrieval R@1 drops from 93.85 to 28.11. The proposed Feature Recovery Module maps degraded encoder features to pristine-aligned representations while keeping the host model frozen, averaging relative gains of 12.5% for retrieval and 20.1% for material recovery across VLM hosts and severity levels.
30
A new open standard locks AI weights to approved hardware
OPAQUE releases Weight Custody Manifest, an open standard keeping AI model weights encrypted until receiving hardware cryptographically attests to builder-specified conditions.
OPAQUE, a confidential computing company, released the Weight Custody Manifest (WCM) standard as a developer-preview specification with a Python SDK and a public test suite of 91 cases. WCM keeps model weights encrypted until the receiving infrastructure proves via CPU/GPU attestation that it matches builder-signed conditions, and decryption access can be revoked later if conditions change. OPAQUE says it ran the attestation exchange on an NVIDIA H100 and on AMD and Intel confidential servers hosted on Azure and Google Cloud, with two independent SDK builds producing identical output across 5,948 files. The public quickstart only exercises protocol logic on synthetic evidence and skips GPU cryptographic verification, and the standard cannot distinguish an authorized key from one physically extracted from hardware.
38
TPMSpy: Validation of Measured Boot Systems by Low-Level Tracing of TPM Usage
Researchers present TPMSpy, a platform-agnostic method validating TPM Measured Boot via low-level tracing, finding inconsistent Linux systemd measurements that break remote attestation and LUKS decryption.
An arXiv paper (2609.05011) introduces TPMSpy, a method that analyzes virtualized system–TPM interactions to independently reconstruct and validate TPM Event Logs without relying on the quoting mechanism, applicable to open and closed systems and demonstrated on Linux and Windows. A longitudinal analysis of Linux systems running systemd versions 245–258 (2020–2025) found wide divergence in Measured Boot usage, undocumented behavioral changes, and no common usage pattern. The authors report inconsistent measurement of user-space systemd services, which prevents reliable remote attestation and LUKS disk decryption on affected systems.
40
TasmScan: Continuation-Aware Taint Analysis for TVM Bytecode with Savelist Abstraction
TasmScan introduces source-free taint analysis for TON smart-contract bytecode, detecting 95.3% of defects with 96.8% precision and 17x speedup.
TasmScan is the first bytecode-level static analysis framework for the TON Virtual Machine, enabling cross-continuation data flow reasoning without source code by modeling savelist semantics through forward register analysis with formal over-approximation guarantees. It lifts bytecode into a typed intermediate representation (TASIR) and performs path-sensitive taint analysis. On a 208-contract benchmark with human-confirmed ground truth it detects 95.3% of defects across five classes at 96.8% precision, and resolves 294,546 dynamic continuation targets with 100% precision across 2,921 registry contracts. It achieves a 17x median speedup over symbolic-execution baselines.
28
[webapps] CubeCart 6.7.4 - Stored XSS
A proof-of-concept stored cross-site scripting exploit targeting CubeCart 6.7.4 was published on Exploit-DB.
Exploit-DB lists a proof-of-concept exploit for a stored cross-site scripting (XSS) vulnerability in CubeCart 6.7.4, a PHP-based e-commerce web application. The listing demonstrates injection of attacker-controlled script that persists in the application, but no exploitation in the wild or CVE assignment is reported in the provided text.
12
You've Got a BUD in Me: Authenticated Reads from Per-Block Write Logs
Researchers propose BUD, per-block write-log digests enabling blockchain validators to serve historical membership and exclusion proofs far cheaper than state-wide tries.
The paper introduces Block Update Digests (BUD), which authenticate each block's write log with predecessor pointers, plus a SuperBUD and exponential hierarchy to turn long unchanged intervals into short proofs. Soundness against adversarial provers and up to f Byzantine validators is proven under archive, attestation, and committee evidence assumptions. Benchmarks show a 50x state-size increase raises the base-BUD path only 1.24x versus 3.1x for in-memory and 69.5x for disk-backed Merkle Patricia tries, with read payloads below 800 bytes and p99 warm verification at 146 microseconds.
14
Survey of filesystem MCP servers: how the "sandboxed filesystem" boundary is enforced (one breach, four defended-by-design)
Survey finds one filesystem MCP server vulnerable to a symlink-based sandbox escape while four others are defended by design.
A survey of five filesystem MCP servers assessed how each enforces its path-confinement boundary against symlink escapes in recursive directory walkers. The iceener/files-stdio-mcp-server is vulnerable to a read-side sandbox escape. The other four are defended by design or make no confinement claim. The author describes the vulnerability class as small but real.
30
What Else Needs Fixing? Exploring Cost-Effective Test-Time Compute for Revision Propagation in Artifacts Generated Through Conversation
A new benchmark shows LLMs reach 68.3-93% accuracy propagating local revisions across conversationally generated artifacts, with parallel-sample selection most cost-effective.
The paper introduces a benchmark for revision propagation: when users request a local change, LLMs must identify dependencies and update all affected parts of an artifact generated through conversation, where context lives in the chat history. Nine revision methods, including sequential reflection and parallel sampling variants, were evaluated on gpt-oss-20b/120b, gpt-5.4-mini, and qwen3.5-9b/27b/122b. Baselines scored 68.3-93% accuracy, and selecting among three parallel samples via LLM-based or medoid selection improved accuracy by 2.2-9.7% as the most cost-effective test-time compute strategy. Code and dataset are released.
30