1/ New model releases seem to climb on benchmarks, but fall short when put in the hands of real users. Today, we launch Agent Arena: a fun and easy way for anyone to discover and compare answers from AI agents, all for free! Yes, even the most powerful pro models. Try it:

4/ We’re releasing this early to gather feedback from the community. If you're training with real human preferences—or building infrastructure to do so—this is for you. Try it:

3/ This work builds on open research and tooling:. - Our collaboration with Nithik Yekollu (who pioneered Agent Arena at @ucbrise) on an arena reward model: - @huggingface (Transformers, TRL, PEFT). - RLHF insights from @natolambert’s open-source book. -.

2/ Arena-RLHF pairs seamlessly with our recently launched Agent Arena: Agent Arena collects high-signal, human preferences in real-world settings. Arena-RLHF lets you fine-tune models from that data.

1/ Today we’re open-sourcing Arena-RLHF, an easy way to RLHF on arena-style human preference data!

RT @Optimism: 2/ Our data partner @openblocklabs breaks down the TVL number across key DeFi categories, generated by 50K+ total participant….

RT @Optimism: 1/The final countdown is on. SuperStacks wraps up today at 23:59PM UTC. You’ve been part of a massive movement showcasing th….

RT @0xTejpal: Had a blast sharing our Agent Arena launch at @convex_dev — the blazing fast DB that powers Love se….

RT @TobinSouth: LM Arena is the gold standard. For agents, we don’t need evals; we need arenas for different tasks. Great take by @0xTejp….

RT @ngundotra: Very cool demo from @openblocklabs using the MCP server to make Raydium pool scanning bot that works!.

4/ Huge thanks to @Solana for hosting the public endpoint at that makes this magic possible!. As OpenBlock's OB-1 reaches more builders, we'd love to collaborate—come build with us. cc @solana @aeyakovenko @SolanaFndn @GitHub 🚀.

3/ You can go to to: .- Query live Solana RPC endpoints and analyze on-chain data.- Look up Solana concepts, APIs & SDKs.- Query Anchor docs and provide tailored dev solutions .- Create GitHub repositories and push production code • Build full-stack.

2/ 🔗 Code & demo.Repo: Build log:

1/ @openblocklabs blockchain agent, added @solana Developer support via MCP. The agent single shotted an open-source Memecoin Signals Bot that monitors Raydium liquidity pools for fresh token listings

RT @0xTejpal: I'll be speaking at @convex_dev's SF office Thursday about OpenBlock's Agent Arena + agent eval best practices. Should be a….

@OpenAI @scale_AI 5/ We're just getting started, but already seeing the power of this approach. Rich, diverse, real-world evaluation data at scale. Try it out, and join our community:.

RT @0xCryptoDypto: This is the future 🔥. Free access to top models (O3, Claude 4, Gemini 2.5 Pro) + credibly neutral evaluation + personali….

RT @0xTejpal: I always find myself debating whether to use 4o or o3, but now with Agent Arena on @openblocklabs I just ask both and pick th….

@OpenAI @scale_AI 4/ With Agent Arena, you can have multiple agents attempt a task and simply pick the best response. No more having to figure out if O3 or 4o is best for your task.

@OpenAI @scale_AI 3/ In exchange for head-to-head comparison data, we are able to offer free inference to top models like O3, Claude 4, and Gemini 2.5 Pro (with tool + MCP access!).

2/ The AI evaluation market just exploded. @OpenAI cut ties with @scale_ai. Enterprise customers are scrambling for alternatives. We're not just another vendor; we're building credibly neutral infrastructure that can't be censored.