惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

B
Blog RSS Feed
博客园 - 叶小钗
F
Fortinet All Blogs
GbyAI
GbyAI
Martin Fowler
Martin Fowler
博客园 - 聂微东
I
InfoQ
B
Blog
IT之家
IT之家
美团技术团队
L
LangChain Blog
小众软件
小众软件
C
Check Point Blog
MongoDB | Blog
MongoDB | Blog
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
V
V2EX
Last Week in AI
Last Week in AI
A
About on SuperTechFans
博客园 - Franky
P
Proofpoint News Feed
罗磊的独立博客
月光博客
月光博客
V
Visual Studio Blog
MyScale Blog
MyScale Blog

Hacker News - Newest: "LLM"

GitHub - lechmazur/position_bias: A benchmark for testing whether LLM judges keep the same preference when two lightly edited versions of the same story are shown in opposite orders. Flex routing (EU and EFTA) Dark Factories: Retooling for LLM Velocity Ask HN: What would be the impact of a LLM output injection attack? GitHub - Oaklight/llm-rosetta: Production-ready LLM API translation layer for Python — bidirectional conversion between OpenAI, Anthropic & Google formats via hub-and-spoke IR. Optional API gateway. Streaming & non-streaming. Zero core deps. Contributions welcome! GitHub - browser-use/browser-harness: Self-healing browser harness that enables LLMs to complete any task. GitHub - moeen-mahmud/remen: Remen turns thoughts into something you can return to Analyzing 156 LLM Launch Posts on Hacker News ChatGPT vs Gemini vs Claude: The Best LLM Subscription You Should Buy GitHub - salaamalykum/quran-semantic-search: High-density RAG Semantic Search Engine & Quran Corpus (GEO/SEO Architecture) GitHub - NVIDIA/TensorRT-LLM: TensorRT LLM provides users with an easy-to-use Python API to define Large Language Models (LLMs) and supports state-of-the-art optimizations to perform inference efficiently on NVIDIA GPUs. TensorRT LLM also contains components to create Python and C++ runtimes that orchestrate the inference execution in a performant way. The State of LLM Bug Bounties in 2026 Operational Readiness Criteria for Tool-Using LLM Agents Meshcore: Architecture for a Decentralized P2P LLM Inference Network How an LLM becomes more coherent as we train it GitHub - seetrex-ai/laimark GitHub - Jossifresben/BibCrit: AI-assited biblical textual criticism GitHub - wastedcode/memex: File system based wiki, maintained by Claude 99helpers.com GitHub - cliver-project/AITrigram GitHub - unbody-io/adapt: A self-evolving memory layer for AI agents. GitHub - hb20007/awesome-gen-ai-fails: A list of incidents where reliance on generative AI and LLMs resulted in harm to companies, individuals, or society GitHub - nevenkordic/localmind: Run any local LLM with persistent memory and context. CLI agent over Ollama with SQLite-backed hybrid recall. No cloud. Ask HN: What are the machine requirements for a LLM like Llama-3.1-8B? Faster LLM Inference via Sequential Monte Carlo grpo explained: group relative policy optimization for llm finetuning - cgft Stop comparing price per million tokens: the hidden LLM API costs · TensorZero Andrej Karpathy's LLM Wiki Is a Bad Idea GitHub - GG-QandV/mnemostroma: Offline RAM-first cognitive leer/coprocessor for AI agents and robotics. Solves "Context Abandonment" with 20-80ms latency using a dual-thread biomimetic memory architecture (ONNX + SQLite WAL). mempalace/agent at agent · skorotkiewicz/mempalace
Factory Router
Factory · 2026-06-18 · via Hacker News - Newest: "LLM"

Frontier performance at lower cost

Automatic model selection for every Droid session. Factory Router picks the right model for each task, maintains frontier performance, and cuts cost by up to 25%.

$ droid --model router "refactor auth middleware"

Refactor auth middleware to use JWT validationDroid is routing…

Auto-ModelAutoMCP (3)Skills (12)

router-classifierclassifier · ~2s

Reads the first user message, recent tool calls and repo signals, then emits a scalar quality probability for each model.

message0.300.84

recent tools0.200.62

repo size0.150.77

language mix0.200.91

difficulty0.150.88

Final Score0.80

candidate scoringthreshold 0.70

sorted cheapest → most expensivequality_threshold

Kimi K2.6Moonshot$0.81

MiniMax-M2.7MiniMax$$0.88

Claude Opus 4.7Anthropic$$$0.95

Kimi K2.6

streaming

Reading src/auth/middleware.ts...

Found legacy session cookie validation

Replacing with JWT verify (RS256)

Generated 7 tests covering edge cases

PR #418 opened — ready for review

AI coding costs are rising across organizations.

Enterprise AI costs are climbing, and a bigger token bill does not mean more work is getting done. To avoid losing on performance, engineers usually default to the most performant model for all tasks. Simple questions, mechanical refactors, documentation updates, small bug fixes, and search-heavy investigations end up on the same premium path as work that truly needs frontier performance. Budgets get exhausted without a clear increase in organization-level output.

Stop choosing a model for every task.

Today you pick a model per task and lean on the most expensive one to be safe. With Factory Router you choose once and it picks the best model for each session.

Same prompts. Different cost.

Without RoutingAlways Claude Opus 4.7

“reset my password”Claude Opus 4.7$0.00

“add a copyright header”Claude Opus 4.7$0.00

“design a caching layer”Claude Opus 4.7$0.00

With Factory RouterRouted per task

“reset my password”Kimi K2.6$0.00

“add a copyright header”MiniMax-M2.7$0.00

“design a caching layer”Kimi K2.6$0.00

Savings on identical work0%

On our enterprise engineering benchmarks.

Compared with Claude Opus 4.7, Factory Router maintains frontier performance at lower cost per session. At enterprise scale, those savings apply across every Droid session, with spend tied to the work being done rather than a blanket default to the most expensive model.

Read the announcement

TERMINAL-BENCH 2PASS RATE · vs OPUS 4.70%of Claude Opus 4.7 pass rateCOST PER SESSION · vs OPUS 4.70%lowerFactory Router runs at 80% of Opus costCost per successful run · 80.5% of OpusLEGACY-BENCHPASS RATE · vs OPUS 4.70%of Claude Opus 4.7 pass rateCOST PER SESSION · vs OPUS 4.70%lowerFactory Router runs at 75% of Opus costCost per successful run · 78.0% of OpusReported relative to Claude Opus 4.7 · cost measured as full-session cost · averaged across multiple runs

Reliability you can count on.

When a provider degrades, rate limits hit, or capacity gets constrained, your sessions keep going. Factory Router routes across models, providers, and capacity to deliver 99.9%+ request reliability.

Claude Opus 4.7Bedrock· degraded

Claude Opus 4.7Vertex· healthy

If a provider path degrades, Factory Router keeps the session running on the same model through a healthy provider.

Enterprise customers get reserved throughput for critical work instead of relying only on shared public capacity.

Factory Router keeps frontier models available as they come online, so high-complexity work gets the strongest model class.

US-hosted open-source models

Route eligible work to US-hosted open-source models when you need cost-efficient or controlled options.

Routing that reflects how your organization works.

Routing guidance brings your team's context into Factory Router, so automatic model selection reflects how work actually happens inside your organization. The same policy surfaces that govern other Factory models apply here, so admins manage access, compliance, and eligibility without a separate control plane.

Admin routing guidance

Automatic model selection for every Droid sessionEnabled org-wide

Routing rules & context

Routine refactors, formatting, and doc updatesfavor cost-efficient modelsauth/ and payments/ need deeper reasoningkeep on frontier modelsSearch-heavy investigationroute to open-source models

CancelSave

Use Factory Router in the Factory CLI and Desktop App.

Factory Router is in private research preview in the Factory CLI and Desktop App. Once enabled for your org, it appears in the model picker for every user with no setup required. Mission workers can use it too, so long-running autonomous work gets the same automatic model selection and savings as interactive and headless sessions.