TechiSeek

VendorsFor Vendors

Memdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard

Hacker News

Memdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard News

Memdoor's team-solve is a workflow that runs five small models on one programming problem: a planner, two ideators, a solver, a manager (review) and a repair worker, each starting from a fresh context and communicating only through files on disk, with a script that runs the samples deciding success. It is the method of GVS5 H (Lee, 2026) implemented as a Memdoor workflow: eight YAML files, no code, no model named.

Practical examples

  • One YAML file per step in steps/, in the format of mario, the engine underneath; requires: is the DAG.

Key takeaways

  • A median of 40 minutes and $0.39 a run; $35.79 for the 50 runs.
  • The loop sees the public samples; the score is on the hidden tests.
  • Parity check (every request byte for byte against their code), hosts, the changes it took and the first run's findings are in NOTES.md; the result on one page is at memdoor.ai/docs/team-solve.
Read More

Recommended

Seoul Biosys Begins Production of Rare Earth-Free White Micro LEDsSeoul Biosys Begins Production of Rare Earth-Free White Micro LEDsHacker News UK Green party drops longstanding opposition to nuclear powerUK Green party drops longstanding opposition to nuclear powerHacker News TCP Use-After-Free: CVE-2026-98432 / CVSS 9.8TCP Use-After-Free: CVE-2026-98432 / CVSS 9.8Hacker News The AI cost of JavaScript frameworksThe AI cost of JavaScript frameworksHacker News AI agent swarms in Minecraft fail to build civilization, drown in bureaucracyAI agent swarms in Minecraft fail to build civilization, drown in bureaucracyHacker News Training: Hard Work Pays off for Data/AI ProsTraining: Hard Work Pays off for Data/AI ProsThenewstack 10 best AI observability tools for monitoring and evaluating agents in 202610 best AI observability tools for monitoring and evaluating agents in 2026Mintlify New Spanish and Turkish Language Models and Updated General Models - Deepgram Blog ⚡️New Spanish and Turkish Language Models and Updated General Models - Deepgram Blog ⚡️Deepgram Rapid `14 Q1 Growth, 23K+ Customers, 45K+ Instances, 4 Clouds (and 10 Regions)Rapid `14 Q1 Growth, 23K+ Customers, 45K+ Instances, 4 Clouds (and 10 Regions)Redis AI frameworks: Definition, types, and how to chooseAI frameworks: Definition, types, and how to chooseZapier Transform and Future-Proof Your Architecture with MACHTransform and Future-Proof Your Architecture with MACHThenewstack Oso Unbundles Security AuthorizationOso Unbundles Security AuthorizationThenewstack Empower Your Team with Policy as CodeEmpower Your Team with Policy as CodePulumi Discovered Stacks: One Place for All Your InfrastructureDiscovered Stacks: One Place for All Your InfrastructurePulumi How a top tier European soccer team sped up their data processing and reduced costs by 50%How a top tier European soccer team sped up their data processing and reduced costs by 50%Modal How Hunch supercharged AI workflows with Modal SandboxesHow Hunch supercharged AI workflows with Modal SandboxesModal Anthropic integration with Modal brings scalable compute to Claude ScienceAnthropic integration with Modal brings scalable compute to Claude ScienceModal Introducing DocChat: GPT-4 Level Conversational QA Trained In a Few Hours - CerebrasIntroducing DocChat: GPT-4 Level Conversational QA Trained In a Few Hours - CerebrasCerebras Cerebras Systems Enables GPU-Impossible™ Long Sequence Lengths Improving Accuracy in NaturalCerebras Systems Enables GPU-Impossible™ Long Sequence Lengths Improving Accuracy in NaturalCerebras How to Run Hugging Face Models Programmatically Using Ollama and TestcontainersHow to Run Hugging Face Models Programmatically Using Ollama and TestcontainersDocker