koca

researcher

[Claude @claudeai]

Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges.

They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.

[Succinct @SuccinctLabs]

Huge congrats to Succinct’s Rahul Dalal, one of the cryptographers behind FlashProve, @AnthropicAI’s provable inference system built on zkSNARKs.

Why: Today, AI runs on a “just trust me” model. Users can’t verify that a response came from the model they chose, or that an attacker didn’t tamper with it along the way. As AI takes on higher-stakes work, that create tension for users, labs, and regulators alike.

What: Researchers from Anthropic, EPFL, MIT, and @SuccinctLabs built FlashProve, a system that lets a model provider commit to a specific model and then prove each response was produced by it, without revealing the weights. Under the hood, it uses zkSNARKs, the same cryptographic technology Succinct uses to secure $13.5B in deposits on Ethereum L2s.

So What: FlashProve proved Llama 4 Maverick, a 400B-parameter LLM, at 110 tokens/sec on a single H100 GPU, with ~19 MB proofs. That puts proving at roughly 70 to 95x the cost of running the model. The previous standard for the same model was ~0.02 tokens/sec.

What’s Next: FlashProve is still a research prototype, but it’s a step change toward the question we asked with Flock earlier this year: how close can proving get to computing? The closer it gets, the less reason there is to release any output without a proof attached.

Congrats again to Rahul and the FlashProve team, and great to see @Starknet, @Mantle_Official (proven by Succinct), and several zkML projects cited.

[Huginn Tech ⚛️ @HuginnStake]

Fibre is now live on Celestia mainnet, and Tensile is live with it.

Tensile is a real-time explorer for Celestia Fibre. It shows each blob as it lands, along with its fee and endorsers, and checks whether validators continue serving the data they endorsed.

Congrats to @celestia on bringing Fibre to mainnet.

https://t.co/UkYWIWZJbh

[koca @kkocality]

@nickwh8te congrats

[koca @kkocality]

@celestia congrats!

[Celestia @celestia]

Introducing Eden: a high-performance EVM chain on Celestia that we use as a sandbox for agentic commerce experiments.

[koca @kkocality]

@altantutar @moremarketsxyz congrats, good luck!

[koca @kkocality]

5️⃣ This isn’t @celestia mainnet doing 3 Tb/s today though. the benchmark used 2 GiB blobs, 1s blocks, 2,000 submissions per block and machine-specific tuning, running the production pipeline from an experimental performance branch. initial mainnet blobs are planned at 128 MiB

what i like here is that 3 Tb/s didn’t come from one magic optimization. every time they made one part fast enough, something else became the limit: encoding, then the chain, then storage

pretty bullish on Fibre after this. definitely one I’ll be watching closely as it moves toward mainnet

if you want to dig deeper into Fibre and the full benchmark:
https://t.co/MLu9a4tssq

[koca @kkocality]

Every transaction @Visa processed in FY2025 would fit into just over two minutes of @celestia Fibre throughput.

Celestia Fibre averaged 3.07 Tb/s over a 143s end-to-end run across 120 validators, enough data for nearly 2B tx/s by Celestia’s estimate

the number is wild, the path to it is more interesting ⤵️

[Celestia @celestia]

A prospective customer with extreme throughput requirements asked how much capacity Fibre can really deliver. So we ran an end-to-end benchmark across 120 validators.

It sustained 3.07 Tb/s of throughput, enough data for nearly 2 billion transactions per second. Every transaction Visa processed in 2025 would fit in just over two minutes.

Full breakdown: https://t.co/VgLyQBnzNi

[koca @kkocality]

4️⃣ Fibre had already hit 1 Tb/s earlier this year, but that was a partial benchmark focused on the networking layer. This time, with those bottlenecks pushed back, they tested the full production pipeline from fresh blob generation and encoding through distribution, storage, validator signatures and finally the onchain commitment

over the full 143s window: 3.07 Tb/s
best rolling 60s: 3.69 Tb/s
best rolling 30s: 4.27 Tb/s

the throughput figures count only blob payloads whose Fibre publications were confirmed onchain. recovery pieces and the traffic used to distribute them were excluded

[koca @kkocality]

@andyyy Bullish

[Nick White @nickwh8te]

Celestia continues to push the frontier of onchain scale with Fibre, hitting the equivalent of 2B TPS and 3x'ing our January bench

Two key differences this time:
1. We ran this at the request of a prospective customer
2. This was an end-to-end production benchmark

Huge result

[koca @kkocality]

@andyyy @therollupco congrats

[koca @kkocality]

@doganeth_en congrats on everything you built these last two years. all the best moving forward!

[The Rollup @therollupco]

new rollup, who dis?

so much more from our team on this side, were having a lot of fun cooking right now.

[koca @kkocality]

@nickwh8te @goodyeartrader @fomo LM

[koca @kkocality]

@andyyy absolute banger