Good read from @BlockworksAdv on the emerging Crypto Inference market. Now there has been one month since this post out .I took a closer look at the latest OpenRouter data and updated the numbers for crypto native inference providers. (Sorry I didn't find any data about @PhalaNetwork on OP)
@VeniceAI now leads with 120.4B tokens/day, followed by @NEARProtocol Near AI (37.7B) and @akashnetAI AkashML (28.9B).
Despite the growing demand for AI inference, crypto native providers still account for less than 1% of OpenRouter's total token volume. Don't forget there are lots of token volume consumed by other giant firm.
As open weight models become increasingly competitive, inference is becoming more commoditized. The real question is no longer just who can provide computation , but who can attract sustainable demand, maintain healthy margins, and compete without relying on token incentives.
Curious to see how this landscape evolves as the market matures.
Note : Date is from OpenRouter and third-party estimates as of Nov 15, 2026.
Evidence of same-block MEV despite there being no public mempool. Almost half of captured MEV happened in the same block via private orderflow. Given Robinhood Chain’s focus on bringing execution fairness here deserves attention. Would appreciate @RobinhoodCrypto's perspective.
Replying to @MEV_X_project
4/5 Private orderflow is highly concentrated: Provider X and unidentified providers carry over 90% of it.
Almost half of all MEV is captured in the block where the opportunity appears, all through orderflow. Open competition took only $7.49M out of $16.71M of captured MEV.
2
76
Claude Opus 4.8 just identified itself as Qwen when asked in Chinese.
Maybe it got distilled a little too hard from Chinese model outputs lol.
Reproduce it:
curl api.anthropic.com/v1/message… \
--header "x-api-key: $CLAUDE_API_KEY" \
--header "anthropic-version: 2023-06-01" \
--header "content-type: application/json" \
--data '{
"model": "claude-opus-4-8",
"max_tokens": 1024,
"messages": [
{
"role": "user",
"content": "你是什么模型"
}
]
}'
1
148
I made Claude Code remember everything across sessions.
Plain text + SQLite beat a 79MB neural embedding model (MemPalace/ChromaDB). No embeddings. No vector DB. Just FTS5 and the right structure.
1
81
Benchmarks & research: github.com/friday-james/zett… (held-out validated, 5 benchmarks, full ablations)
Skill: github.com/friday-james/zett… (MIT)
Built with Claude Opus. The retrieval problem is solved. The future is structure, not similarity.
1
50
a Claude Code skill that scans your npm and pip dependencies for vulnerabilities, typosquatting, and supply chain attacks.
github.com/friday-james/depe…
1
42
I built a tool that can audit and iteratively edit your code until it reaches a defined quality bar — think automated review + fix loops (a “Ralph loop” style workflow).
Designed for individual devs who want strong guardrails without slowing down.
Repo: github.com/friday-james/let-…












