AI/ML●●●Banger
97% on SWE-bench Verified with subscription-token agents
97% on SWE-bench Verified with full artifact transparency, not just a score claim.
Big BrainZero to One
kimjune01
2019d ago
Back to browse
Codex context bloat? 87% avg reduction on SWE-bench Verified traces
by george_ciobanu·Apr 24, 2026·10 points·2 comments
AI Analysis
●●SolidBig BrainNiche Gem
Transparent proxy cuts Codex context tokens by 87% via working memory.
Strengths
- •Benchmarks show massive token reduction on SWE-bench traces.
- •Zero code changes required, sits between Codex and OpenAI.
- •Open source TypeScript implementation includes replay testing suite.
Weaknesses
- •Tied specifically to Codex, brittle if upstream API changes.
- •Working memory logic is opaque, risk of losing critical context.
Category
Target Audience
Developers using OpenAI Codex for software engineering tasks
Similar To
Cursor · Continue · LangChain Memory
Post Description
If you had to build a context window manager in 24h, would you stick to the existing model or come up with something better?
Here's what I did:
1. Built a proxy that intercepts Codex's calls to OpenAI and rewrites them on the fly.
2. Replayed 3,807 rounds of SWE-bench Verified traces through it: avg prompt 44k → 6k tokens (-87%).
3. Posted it to HN to get the next reduction applied to my confidence interval — starting with the inevitable "How about accuracy?"
npx -y pando-proxy · github.com/human-software-us/pando-proxy
Similar Projects
AI/ML○Pass
All the LM solutions on SWE-bench are bloated compared to humans
Twitter thread with a chart; not a product or tool.
lieret
103mo ago
Developer Tools●●Solid
A open-source, local trace viewer for Claude Code and Codex sessions
Chrome DevTools for Claude Code sessions when LangSmith drops local tool calls.
Solve My ProblemNiche Gem
MediumD
101mo ago
Developer Tools●Mid
Salacia – The First Runtime OS for Agentic Coding
Fault-localization scaffolding for AI agents; claims 93% top-5 recall, but Cursor/Cline already integrate similar.
Big BrainBold Bet
alfredhua
203mo ago
AI/ML●●Solid
PrismoDev – local CLI for finding token waste in Claude Code/Codex
Finally, a tool that tells you why your AI coding bill is exploding.
Solve My ProblemBig Brain
shanirshad
2124d ago
Developer Tools●●Solid
We achieved 72.2% issue resolution on SWE-bench Verified using AI teams
They split responsibilities across isolated agents (engineer, reviewer, manager) that get real shell access and independent filesystems, which makes failures traceable and lets you tune model capacity per role. Hitting 72.2% on SWE-bench Verified with no benchmark-specific tuning is an impressive empirical result — interesting architecture and strong evidence — though the security and long-term reliability of autonomous shell-executing agents remain the big open questions.
WizardryBig Brain
NBenkovich
204mo ago