this moves blockchain from simple automation → toward intelligent execution.
instead of if this happens, do that, the system can now reason through multiple possibilities before acting.
English
melophile arc
1.5K posts
melophile arc
@biotechdrops
cooking something
Bull Run, VA Katılım Nisan 2025
213 Takip Edilen81 Takipçiler
~ a defi protocol could adjust strategies based on market conditions
~ an ai agent could manage treasury allocations automatically
~ onchain systems could analyze risks before executing transactions
• app could respond differently depending on user behavior or live data
English
intelligent smart contract execution is about making smart contracts more flexible and aware instead of only following rigid pre-written rules
traditional smart contracts can only execute exact instructions
let's dive in
melophile arc@biotechdrops
genlayer is optimized for onchain and ai-native applications because traditional blockchains were never designed for ai-level reasoning and execution most chains are built for fixed logic but ai works differently it processes context, adapts to inputs, and generates decisions
English
TwinRouterBench makes this even more interesting
instead of testing clean prompts only, it benchmarks real messy multi-step agent workflows with dynamic execution.
ai routing is moving from “always use the biggest model” → “use the right model for the right step.”
English
- up to 82% lower api cost
- 93.4% task pass rate
- 75/100 swe-bench tasks solved
- ~53% cheaper than opus-only setups
English
most ai agents today waste money by sending every task to expensive frontier model
uncommonroute solves this by acting as a smart local llm router
instead of using one model for everything, it automatically picks the cheapest model that can still handle the task properly
English
current routing benchmarks still test single prompts in isolation
TwinRouterBench stands out it pushes evaluation closer to real agent behavior
the separation between static supervision and dynamic SWE-bench execution is a really solid approach
Yuhang Yao@yuhang_yao
Excited to share that TwinRouterBench has been accepted to the #RLEval Workshop at #CAIS2026 🎉 As LLM apps become long-horizon agents, one request can trigger many model calls across planning, tool use, retrieval, coding, and verification. That makes per-step LLM routing a core infrastructure problem: sending each call to the cheapest sufficient model without breaking downstream success. TwinRouterBench introduces: ⚡ Static track: 970 router-visible prefixes from 520 instances across SWE-bench, BFCL, mtRAG, QMSum, and PinchBench 🚀 Dynamic track: live SWE-bench Verified evaluation with official task resolution + realized API spend Key result: a router trained on static labels achieves comparable SWE-bench resolve rate while cutting API cost by ~53% vs. an unrouted Opus 4.6 baseline. Paper: arxiv.org/html/2605.1885… Code: github.com/CommonstackAI/… Dataset: huggingface.co/datasets/Amorp… Website: commonstackai.github.io/TwinRouterBenc… #LLM #AgenticAI #LLMRouting #Benchmark #SWEBench
English
Great to see TwinRouterBench accepted to the #RLEval Workshop at #CAIS2026!
Per-step routing is quickly becoming essential infrastructure for agentic systems: each planning, coding, retrieval, and verification call should use the cheapest sufficient model without hurting final task success.
Proud to open-source TwinRouterBench and contribute a practical benchmark for this problem.
Yuhang Yao@yuhang_yao
Excited to share that TwinRouterBench has been accepted to the #RLEval Workshop at #CAIS2026 🎉 As LLM apps become long-horizon agents, one request can trigger many model calls across planning, tool use, retrieval, coding, and verification. That makes per-step LLM routing a core infrastructure problem: sending each call to the cheapest sufficient model without breaking downstream success. TwinRouterBench introduces: ⚡ Static track: 970 router-visible prefixes from 520 instances across SWE-bench, BFCL, mtRAG, QMSum, and PinchBench 🚀 Dynamic track: live SWE-bench Verified evaluation with official task resolution + realized API spend Key result: a router trained on static labels achieves comparable SWE-bench resolve rate while cutting API cost by ~53% vs. an unrouted Opus 4.6 baseline. Paper: arxiv.org/html/2605.1885… Code: github.com/CommonstackAI/… Dataset: huggingface.co/datasets/Amorp… Website: commonstackai.github.io/TwinRouterBenc… #LLM #AgenticAI #LLMRouting #Benchmark #SWEBench
English
@SuiInsiders intrested
I have prev experience working late night, so adapting schedules isn’t really an issue
looking forward to hear
English
We’re hiring 👤
Can you stay awake from 1 AM to 6 AM?
If yes, I may have a remote opportunity for you paying up to $100/hour 🤑
We’re looking for a few reliable people who can work around 6 hours daily.
Drop a reply below 👇
And don’t forget to check your DMs later ✉️
English
x is removing the x communities feature on may 30th
so unfortunately the grad community won’t be available there anymore
but the community isn’t going anywhere
you can join us on the gradient_hq discord
discord.gg/gradientnetwork just hop in
English
melophile arc retweetledi
i was the fourth member of lounge (**wink wink** 😏)
anyways join @Gradient_HQ discord as well here:
discord.gg/gradientnetwork
see you guys there!
rw ./@gradientintern
Hello everyone. X is removing it’s X communities feature on May 30th I started building this community in January 2025 and it’s been a wonderful experience to meet nearly 5,000 of you inside of our Open Intelligence Lounge @Gradient_HQ will continue to update it’s research efforts on main page and on Discord with other activities as well: discord.gg/gradientnetwork Let’s stay in touch and there will be more to come!
English
instead of blockchains only verifying transactions, genlayer pushes them toward executing intelligence itself.
the goal is to create an environment where ai becomes a native execution layer for web3 applications, not just an external tool connected through APIs
English
genlayer introduces an architecture where intel app can operate directly onchain instead of relying on offchain systems.
this enables
< autonomous ai agents
< ai-driven smart contracts
< real-time reasoning systems
< adaptive defi strategies
< programmable decision making
English
genlayer is optimized for onchain and ai-native applications because traditional blockchains were never designed for ai-level reasoning and execution
most chains are built for fixed logic
but ai works differently it processes context, adapts to inputs, and generates decisions
melophile arc@biotechdrops
Exciting to see the momentum around @GenLayer The bradbury testnet is already being secured by a strong validator Glad to see builders, operators and infrastructure teams coming together this early @encapHQ @nansen_ai @NodesGuru @SenseiNode @stakeme_pro
English
grad love test score error 404
found: love with @Gradient_HQ
check yours grad404.vercel.app
rich ppl go all the way, grad ppl go all the way
thanks to @realsirandrew
realsir ./@realsirandrew
think open, u know 404: grad404.vercel.app rich ppl go all the way, grad ppl go all the way @Gradient_HQ ./
English
ai isn’t only about powerful models anymore
the real thing is using the right model for the right task at the right cost
that’s why <routing> is the most important ai layers
exactly why stuff like uncommonroute from @commonstack_ai makes sense
github.com/CommonstackAI/…
Commonstack@commonstack_ai
Fraction of the bill. Same results. Fully local, open source, works with any client. Just > pipx install uncommon-route github.com/CommonstackAI/…
English
Fraction of the bill. Same results.
Fully local, open source, works with any client.
Just > pipx install uncommon-route
github.com/CommonstackAI/…
English