Liam Nerd

Bay Area AI Benchmarks, shared daily.
Follow for god-mode workflows, real hacks & faster shipping.

[Liam Nerd @LiamAITools]

Have you tested the codex token-saving tricks everyone keeps recommending? 🫣Do they actually work?
📒I found ~20 of them from github, x, youtube, and short-video tutorials. (Caveman. Ponytail. MarkItDown. AGENTS.md. /clear. prompt caching...)
📢some claim 80% savings, but is that actually true?🧐

so I’m running a 20-day test:
same task style.
real screenshots.
usage numbers when available.
kept / killed verdict.

The survivors become a step-by-step handbook at the end. Follow if you want REAL results.🤑

#Ai #ClaudeAI #Codex #Chatgpt #Tokenusage

[Liam Nerd @LiamAITools]

@_lexers @Accio_official 1k+ stars fast for a reason. Curious how much the harness vs base model moves that 61.7%.

[Liam Nerd @LiamAITools]

@systemdesignone Needed a commerce counterpart to SWE-bench / OSWorld. Scoring environment state, not the model’s self-report, is the design choice that matters.
https://t.co/FZtsyXC2vS

[Liam Nerd @LiamAITools]

@togethercompute @Zai_org Budget-per-task is the metric that sticks.

[Liam Nerd @LiamAITools]

@ZixuanLi_ Delay for framework support is the adult move.

[Liam Nerd @LiamAITools]

@HARNESSROUTER Stop rewriting integrations for every agent runtime.

[Liam Nerd @LiamAITools]

@hxiao @ZixuanLi_ Agent skills matter more than the launch table.

[Liam Nerd @LiamAITools]

@Raullen That’s how you keep MoE quality.

[Liam Nerd @LiamAITools]

@MiaAI_lab @Kimi_Moonshot @MiniMax_AI Lightweight open variants are the new default.

[Liam Nerd @LiamAITools]

@jundotkim N-gram offload + MTP actually moved tok/s.

[Liam Nerd @LiamAITools]

@karanC_12 Quality without latency is still half the product.

[Liam Nerd @LiamAITools]

@ItsmeAjayKV Local Flash-Next just became more usable.

[Liam Nerd @LiamAITools]

@Zai_org Stealth window over. Weights out.

[Liam Nerd @LiamAITools]

@UnslothAI @Zai_org Local Flash club just got bigger.

[Liam Nerd @LiamAITools]

@ArtificialAnlys Agent benches need this kind of honesty filter.

[Liam Nerd @LiamAITools]

@opencode Free windows measure real demand better than benches.

[Liam Nerd @LiamAITools]

@MiaAI_lab Time to measure, not just screenshot the launch card.

[Liam Nerd @LiamAITools]

@business OpenRouter free windows rewrite adoption curves.

[Liam Nerd @LiamAITools]

@testingcatalog Now the nameplate is public.

[Liam Nerd @LiamAITools]

@vllm_project @Alibaba_Qwen Serving stack showed up with the weights.