OpenBlock
@openblocklabs
Followers
5K
Following
214
Media
146
Statuses
441
We're building state-of-the-art code and research agents to democratize the means of software production.
San Francisco, CA
Joined November 2022
Introducing OB-1: the new #1 coding agent on Terminal Bench. After a year of R&D, our agent now outperforms Codex and Claude Code. Early access is rolling out to waitlist users now.
19
17
132
So much fun hosting the CMU builder night tonight; packed with demos, energy, and great people. Gave a sneak peek of @openblocklabs' upcoming CLI agent, OB-1! s/o @waynesutton @convex_dev for the space :)
1
4
18
Has anyone tried OB-1? @openblocklabs: https://t.co/HtYBZGXMMw Details:
openblocklabs.com
Experience the future of autonomous AI. OB-1 applies our reinforcement learning research to create agents that think, learn, and evolve.
Introducing OB-1: the new #1 coding agent on Terminal Bench. After a year of R&D, our agent now outperforms Codex and Claude Code. Early access is rolling out to waitlist users now.
40
3
50
The team is planning to support EVM / Solidity - reach out
Introducing OB-1: the new #1 coding agent on Terminal Bench. After a year of R&D, our agent now outperforms Codex and Claude Code. Early access is rolling out to waitlist users now.
5
9
103
We’re hiring! Coding agents are one of the hardest and most important problems. Right now, the same tools used for “vibe coding” are being asked to shoulder the weight of real-world engineering across diverse developer ecosystems; and too often, they fall short. We’re a lean,
Introducing OB-1: the new #1 coding agent on Terminal Bench. After a year of R&D, our agent now outperforms Codex and Claude Code. Early access is rolling out to waitlist users now.
6
3
31
We built the #1 Coding Agent on Terminal Bench. Now, we're looking to partner with a few exceptional teams to build the most advanced agents for their enterprise workflows. Join intimate dinners in San Francisco with other technical leaders who are pushing the boundaries for
4
2
12
Taking #1 on Terminal Bench against teams like Anthropic, Block, and Warp is impressive. Congrats to the OpenBlock team!
Introducing OB-1: the new #1 coding agent on Terminal Bench. After a year of R&D, our agent now outperforms Codex and Claude Code. Early access is rolling out to waitlist users now.
6
4
29
4/ We’re onboarding early partners to try our agent. Early access: https://t.co/ubkvCmotUR Join us: https://t.co/KHh71oaCKo Discord:
1
0
9
3/ Frankly, most benchmarks suck. While others lack diversity (~50% of SWE-Bench is Django), Terminal Bench curates long-horizon tasks that test real-world agency. That’s why we chose it. https://t.co/sDUWZ4Hn4v
2
2
10
2/ Other coding agents deliberately nerf capabilities to save costs. OB-1 doesn’t. Instead, it fuses intelligence across multiple agents with persistent long-horizon memory, delivering a far more reliable coding experience. Sign-up for early access: https://t.co/ubkvCmotUR
2
1
16
We started over a year ago with text-to-SQL agents. Since then, we’ve learned a ton about building agents –– it’s wild to see how far a lean team can push. Now we’re applying this tech into high-value downstream tasks: everything from data analysis to Ethereum development and
1/ OpenBlock just secured #2 on Terminal Bench with our frontier coding agent, OB-1. Our result illustrates how specialized agents built by small teams can outperform the monolithic systems of big labs.
1
1
11
4/ We look forward to working with developer ecosystems to apply our agents to real-world tasks. Stay in touch: Twitter: @openblocklabs Telegram: https://t.co/AJcaD6AmGA Discord:
0
0
1
3/ We built OB-1 to mimic how a capable engineer works: – Persistent memory blocks to track goals, todos, and notes – Adaptive timeouts that keep the agent moving – Editable reasoning so it can revise and self-correct – Direct terminal execution to finish tasks cleanly
openblocklabs.com
Experience the future of autonomous AI. OB-1 applies our reinforcement learning research to create agents that think, learn, and evolve.
1
0
2
2/ Terminal Bench is a rigorous evaluation of agent capabilities, requiring end-to-end task completion in real-world terminal environments, including debugging, configuration, compilation, and validation. https://t.co/sDUWZ4Hn4v
tbench.ai
A benchmark for terminal agents
1
0
2
1/ OpenBlock just secured #2 on Terminal Bench with our frontier coding agent, OB-1. Our result illustrates how specialized agents built by small teams can outperform the monolithic systems of big labs.
6
3
39
So much fun presenting our @openblocklabs Agent Arena at @Vercel HQ! Also got to see an insanely cool v0 launch that’s coming very soon 🔥 s/o @sonofalli @rauchg for hosting!
1
3
26
I'll be speaking this Friday at @ethereumhouseSF on how to train frontier AI agents for blockchain tasks. More details in the thread:
1/ AI is coming to Ethereum. And Ethereum is ready to be its backbone. This Friday, @EthereumHouseSF hosts a gathering of the builders shaping AI x Ethereum — in the city that’s the global capital of AI. Here’s why it’s important👇
1
4
25
Our EVM Text-to-Transaction environment is now available in the latest Atropos upgrade!
Atropos v0.3 is now out! Our RL Environments framework has seen a lot of upgrades since v0.2 - some highlights: - Atropos can now be used as a benchmarking and evaluations framework by @rogershijin, with our first external benchmark, Reward-Bench 2! - Added the Reasoning Gym,
0
1
5
Grok 4 is now live on Agent Arena! Try it out with your favorite tools and MCPs, from web search to scanning on-chain data. Inference is 100% free when you rate model outputs.
0
1
17