Eran Sandler (@erans) header
Eran Sandler (@erans) profile picture

Builder, operator and investor. Infra, AI, and product nerd. Opinions are my own. Checkout @lunaroute_ai (https://teapaw.org/t.co/YlZrgrIUKq) to stop caring about token limits

Joined April 2007
Photo by Eran Sandler (@erans)Photo by Eran Sandler (@erans)Photo by Eran Sandler (@erans)Photo by Eran Sandler (@erans)Photo by Eran Sandler (@erans)Photo by Eran Sandler (@erans)Photo by Eran Sandler (@erans)Photo by Eran Sandler (@erans)Photo by Eran Sandler (@erans)Photo by Eran Sandler (@erans)
Eran Sandler retweeted
Those 100B DeepSeek 4.1 Flash tokens, were made on @lunaroute_ai :) Stop paying by the token. Free trial or talk to me.
Jesse Vincent ran 100 billion tokens in a month. He also saw AI code reviews stretch from 3 rounds to 80. He discusses agent harnesses, review loops, and knowing when to stop fixing bugs.
1
3
3
100
Finally got my @FrameworkPuter 13 Pro. That's one solid machine. And the keyboard is like the older Thinkpads! Love it.
1
2
128
Eran Sandler retweeted
Hot off the press - @slashlast30days v4 - the biggest release since v3. Last30Days - the world's #1 agentic search skill with ⭐68K stars. v4 finds what isn't in the news yet. It has "secret jobs page signal search" and "patent search" built in. Plus quote lines! the Reddit/TikTok/YouTube Comments look so much cleaner. WATCH THE DEMO VIDEO! NEW: 🔬 Patent secret signals: reads a company's new patent filings and flags only the ones nobody has reported 💼 Jobs page secret signals: reads the company's own careers page and spots the roles that give away what's next 💬 Quote lines: the best Reddit, X, YouTube and TikTok comments now sit right under the story they back up 📊 Polymarket odds in one clean line, with how much they moved 📈 Yahoo Finance: the stock's last 30 days vs the S&P 500 🏢 Knows who you mean: "Apple" finds the company, its site and its ticker on its own 🟠 Reddit can search as your own logged-in account 📰 Digg now covers every topic, not just AI and tech FIXED: 🔗 Links stay clickable in Claude Code 🔢 No more vote counts on every line 🎵 YouTube works for artists with a big back catalog 👥 Same-name companies stay out of your reports 98 commits from 27 people since the last release.
33
14
5
131
12,709
Eran Sandler retweeted
We launched opencomputer.dev (again!) on Friday and signups have been BUZZING! As a reminder: - Open Source, Self hostable alternative to Devin - 3.3x faster than Devin - ~local speeds, sometimes faster - Self hostable in AWS, GCP and Azure - Works with your subs! Check the repo out: github.com/diggerhq/opencomp…
Introducing opencomputer.dev An open source, faster Devin alternative you can self-host 0:23 Why it's 3.3x faster 1:27 One VM is all you need 2:01 Fully self hostable 2:16 Faster + cheaper at scale 3:27 Durable + secure 3:51 Star us on GitHub!
3
3
9
758
I would argue that smartest model always mean better or faster. You need the right type of results and paying more doesn't guarantee that. Nor it guarantees faster if you don't get the right results. It makes you experiment less because of cost. More expensive models have a place and use and you shouldn't be paying by the token for most of your tasks anyway. Check @lunaroute_ai to see how we do it.
Coding tasks are saturated. So why is everyone still paying for the most expensive model from Anthropic and OpenAI? a16z's Martin Casado has the answer. Casado, General Partner at a16z and investor in Cursor: 1. "I've automated 95% of my coding" may be true 2. "So I can use a cheaper model" doesn't follow 3. If two startups compete, neither picks the dumber model 4. In a competitive market you always pay 10x for the smartest one 5. So when spending slows, prices hold and usage spreads beyond code This is why the frontier labs keep their pricing power, even in a slowdown. Full breakdown here: theaiopportunities.com/p/ins…
1
200
Eran Sandler retweeted
Replying to @lunaroute_ai
@lunaroute_ai added inkling-2-small earlier today, we're currently running our design harness across briefs, variants and skills and will soon share the results. While it's running we got a glimpse of the early results and boy this one seems promising for its size.
1
2
1
3
217
Eran Sandler retweeted
Introducing opencomputer.dev An open source, faster Devin alternative you can self-host 0:23 Why it's 3.3x faster 1:27 One VM is all you need 2:01 Fully self hostable 2:16 Faster + cheaper at scale 3:27 Durable + secure 3:51 Star us on GitHub!
51
47
16
298
25,215
We are testing @thinkymachines Inkling 2 Small at @lunaroute_ai and so far its looking pretty good. I never take benchmarks at face value so throwing my own workloads at it and its doing pretty good for its size. Much better than the full Inkling 1. It's slightly bigger than Beam, half of Muse Spark Open.
1
1
14
821
Think about all the optimizations we are doing these days around model cost. It feels like we are optimizing for the wrong thing. This is why we are taking a different approach at @lunaroute_ai w/ private (true ZDR), US based servers and fixed cost inference!
4
193
Why summarize an AI agent's conversation when the entire history is on disk? I built a Pi extension that starts fresh when context fills up, while keeping the full history searchable. Cleaner context, no lossy summaries. eran.sandler.co.il/post/2026…
1
2
6
229
Eran Sandler retweeted
If you're in need of dedicated inference that will cost you a fraction of what you're paying OAI or A/ talk to @erans and @lunaroute_ai . You'd be amazed how big of unlock it'll be for your teams
We will see more and more companies doing this. Not paying by the tokens enables Continuous AI for doing all sorts of things. If you can't afford 53 B200 nodes talk to us at @lunaroute_ai - we can get you private fixed cost inference and more than just GLM 5.3 for much less!
1
1
2
309
Not being token scarce is a HUGE unlock for everyone. Combine that with the fact you can now run Continuous AI for sweeping tickets/stories, adversarial code reviews or anything else you never end up doing because of time or cost. Now you can - with @lunaroute_ai #continuousai
120
We will see more and more companies doing this. Not paying by the tokens enables Continuous AI for doing all sorts of things. If you can't afford 53 B200 nodes talk to us at @lunaroute_ai - we can get you private fixed cost inference and more than just GLM 5.3 for much less!
BREAKING: Revolut moved ~2,000 engineers onto self-hosted GLM for coding and pushed Claude Code and Codex to the exception list. Per David Abitbol, who leads AI ops there: 56 Nvidia B200s, a blended 30 cents per million tokens, roughly a tenth of what they paid Anthropic and OpenAI. Engineering lead time improved 70%. Here's the part that matters: the gains showed up when tokens stopped being rationed. Engineers quit saving AI for the hard tickets and pointed it at everything. Frontier models become the exception, open weights become the default. That's a pricing problem for every lab heading to IPO. h/t @OliverMolander Would you move your team off frontier APIs if the bill dropped 90%?
2
1
1
8
658
I keep thinking about all the useful work we don't do simply because it's not urgent. Slow queries, security checks, research, customer feedback, ideas we never test. That's the work I want agents doing in the background, and a big reason we're building @lunaroute_ai the way we are.
Important doesn't always mean urgent. Performance, security, research, data analysis, untested ideas. There's always useful work we never get to. Continuous AI lets agents tackle it in the background. LunaRoute's fixed-cost inference and background scheduling make that practical. lunaroute.com/blog/continuou…
3
139
Eran Sandler retweeted
I blame @erans for creating @lunaroute_ai
1
1
3
101
Checkout @opencomputerHQ - they have been cooking some interesting things lately!
Introducing @computer. Claude Code & Codex in the cloud, as fast as your local machine. Sometimes even faster! Try it today via Slack, byo sub supported. opencomputer.dev
4
1
1
6
539
If an AI model refuses to help you investigate a vulnerability in your own code, that's a problem. Security teams need to explore, test, fail, and try again. We're working on giving them models that can actually help, with verified access and without worrying about the token meter. Curious to see how far we can push this.
Security teams shouldn't hit AI refusal walls when trying to protect their systems. Introducing LunaRoute Cybersecurity & Cyberdefense. 🔍 Vulnerability research & testing 🛡️ Threat analysis & defense Reduced-refusal models. Verified access. No token meter. lunaroute.com/blog/introduci…
2
1
326
Look at the throughput that Vera Rubin can do even on a large model as Kimi K3, it's insane: teapaw.org/SemiAnalysis_/status/2… Remember what I said, even Vera Rubins have the same amount of HBM memory (at the current version) as B300 - so bigger models (think Kimi K3.1, Qwen 3.8 Max and future models) that are close or over 3T params are no longer capable of running on a single node, unless quantization and other tricks are deployed.
BREAKING NEWS🚨: NVIDIA just submitted its first Rubin NVL72 Kimi K3 submission, and it can reach over 850 tok/s/user, even at 2.8 Trillion parameters, with over 3x higher throughput on @vllm_project! Great work by the NVIDIA team, and excited for Rubin Software 1.0 GA coming soon!
3
217