A different kind of insider. Scanning Web3 for AI that break the norm.🔥
Metaverse
Joined April 2025
- Tweets1.7K
- Following1.4K
- Followers2.8K
- Likes6.9K
🤖 Artificial Analysis just launched Coding Agent Index v2.0. The index evaluates models together with their agent harnesses, across 290 tasks covering software engineering, terminal use, and repository Q&A, with the three benchmarks weighted equally.
Top 4:
🥇 Claude Code + Opus 5.5 (max): 61.7
🥈 Claude Code + Sonnet 5.5 (max): 59.3
🥉 Muse Code + Muse Wonder Full (max): 58.9
4️⃣ Claude Code + Fable 5.1 (max, with fallback): 57.6
Next up: Codex + GPT-6 Astra (55), Codex + GPT-6.1 Sol (54), Antigravity CLI + Gemini 4 Argon (53).
What changed from v1.5:
→ FrontierCode v1.1 (from Cognition, private test set, graded on whether a maintainer would merge the change) replaces DeepSWE
→ SWE-Atlas-QnA grading is stricter: the mean score fell from 58.8% to 57.1%, with 14 of 21 configurations moving down
The most interesting part is the cost frontier, four configurations that are optimal on score versus cost:
→ DeepSeek V4.1 Flash: 47.3 at $0.39 per task
→ GPT-6.1 Sol: 54.1 at $1.50 per task
→ GPT-6 Astra: 54.7 at $7.04 per task
→ Opus 5.5: 61.7 at $11.42 per task
Quick math: GPT-6.1 Sol delivers about 88% of Opus 5.5’s score at less than 1/7 of the cost, and DeepSeek V4.1 Flash delivers about 77% of the score at roughly 3% of the cost. Top score means top price, but if you optimize for cost, the gap is smaller than the leaderboard suggests.
📊 SemiAnalysis just limit-tested nearly every AI subscription plan (Anthropic, OpenAI, Meta, SpaceXAI, Cursor, Cognition, the Chinese labs...) and converted the results into API-equivalent value.
Key findings:
→ At the mid-tier (Opus 5.5 vs GPT-6.1 Sol), Claude plans offer roughly 4-5x more API-equivalent value than OpenAI’s plans
→ OpenAI just cut the value of its $200 plan in half and added a new $500 tier, which offers only 21% more Astra than the old $200 plan ($200 plans bought before the cut keep the old limits until Oct 29)
→ After the cut, OpenAI’s Pro $100/$200/$500 plans all offer the same tokens per dollar
→ At the top tier (Fable 5.1 vs GPT-6 Astra), limits are fairly similar, but Fable can only be used for up to 50% of a plan’s limit
On the economics:
→ Subscriptions are only ~10% of Anthropic’s revenue but consume over 40% of inference compute
→ The two labs chose different ways to cut subsidies: Anthropic gradually lowers value on pricier models, while OpenAI cut everything at once
→ When API prices drop, plan limits often don’t rise to match, so a plan’s real value can quietly shrink (OpenAI kept Sol limits unchanged, and value fell about 30%)
Takeaway: you can’t say a plan is “worth $X.” You have to consider the full (plan, model, workload) combination.
Two major AI agent signals on Oct 21:
@nvidia OpenShell - hardware-enforced boundaries for agents (BlueField-4 Sentry can quarantine in ms). Jensen: “An agent cannot govern itself.”
Walmart opens the door with Agent Gateway (MCP) + Agent Profiles + spend caps — while Amazon blocks agents.
Infrastructure + commerce are racing ahead.
xitter.cf/nvidia/status/21128472…
We’re entering the “permissioned agent” era.
NVIDIA is building the kill-switch layer enterprises will demand. Walmart is building the consumer control layer (profiles + limits). Amazon’s wall approach may not scale.
Winners will be those who treat agents as semi-trusted actors with hard guardrails, not fully autonomous employees. Governance > pure capability right now.
Anthropic @AnthropicAI just dropped Claude Duet - a proactive AI agent that works while you step away.
It monitors your apps, sends a morning brief of what actually needs attention, and keeps going on a cloud browser. Built on Opus 5.5 with strong judgment on when to escalate.
Big step toward true always-on agents for pros & SMBs.
claude.com/product/claude-du…
The real differentiator isn’t autonomy - it’s the approval gates + safety classifiers. Duet never sees passwords and must check with you before money/docs/deletes.
This is the right design pattern as agents move from chat to action. Early preview, but the bar for “trusted proactive agent” just got higher.
Would you hand over your inbox + calendar yet?
xitter.cf/claudeai/status/211298…
. @Mastercard just opened Early Access to Agent Pay for Machines.
AI agents can now: • Discover paid digital services
• Negotiate & pay autonomously
• Stay inside human-set budgets
Built on x402 protocol + Base + Coinbase infrastructure.
Live demo: one agent ran a full ad campaign with 9 automatic payments across 2 merchants on a single approved budget.
Pilots already done with @coinbase @base @Cloudflare @BVNKFinance and others.
This is the payment rails for the machine-to-machine economy.
xitter.cf/Mastercard/status/2112…
AI agents can find what they need. Now they can pay for it, too.
Today at @Cloudflare Connect, Mastercard showed how Agent Pay for Machines lets an AI agent find and pay for digital services on behalf of a business, while sticking to the budget and rules that business sets.
Following pilots with @Cloudflare, @Ant_Intl , @BVNKFinance, @AIsaOneHQ, @coinbase, @base and @hypepartners, we're opening Agent Pay for Machines Early Access.
Now we want developers to put it to the test. 👇
.@walt_io (ex-Wallet in Telegram) just dropped AI agent trading for perpetual futures.
You can now connect ChatGPT, Claude, Grok or their own Mira agent and let it:
• Analyze markets in real time
• Open/close longs & shorts
• Manage TP/SL & risk
• Monitor positions 24/7
All inside Telegram.
The agent only gets a dedicated account you fund - it cannot withdraw.
Execution starts OFF by default.
This is the first time mainstream users can hand high-leverage crypto trading to frontier models with one chat.
150M+ potential users. The experiment just went live.
xitter.cf/walt_io/status/2112540…
Your AI perps assistant on Walt Street
Walt is opening up a new way to trade: for the first time in Telegram, an AI agent supports you at every step. You can now get the picture on the market, open a position and check your portfolio risk in one chat, without switching screens. Your agent works with perpetuals and places every trade only once you confirm.
🧠 Connect the AI you already use
Connect ChatGPT, Claude, Grok, Cursor or any other AI agent. Or use @trymira right in Telegram.
⚡️ Your agent gets its own account
You decide how much goes in. The agent can’t withdraw funds, and you can disconnect it at any time.
✅ How to get started:
• Open Walt
• On the home screen, tap “Create AI agent account”
• Pick your agent and connect it
This is the third of five Walt Street updates. Earlier, we showed you how to keep your favorite assets within reach and how to pay less in fees.
#WaltStreet
AI isn't just about building the next big thing. It's also about modernizing the systems the world already runs on.
From COBOL to Fortran and Visual Basic, OpenAI's latest advances aim to help developers modernize legacy systems while preserving decades of critical business logic.
The future of AI-powered coding isn't just about writing new code. It's about making existing systems better.
xitter.cf/OpenAI/status/21123103…
Our latest model advances bring stronger capabilities for building and migrating software in low-resource programming languages such as COBOL, Fortran, and Visual Basic.
These languages power software that banks, governments, and businesses depend on. Progress here means understanding decades of business logic and helping developers modernize those systems while preserving what works.
We’re building these capabilities around the challenges developers and enterprise customers bring us. If your team is working on these systems, tell us which languages and workflows we should tackle next.
openai.com/index/low-resourc…
Anthropic just committed ~20% of its shares to a Public Benefit Fund.
- Starts at 5%
- Up to an extra 2% for every 50% increase in company value
- Co-founders already pledged 80% of their equity
- ~half of employees are donating through a matching program
While traditional AI labs lock value into controlled funds, the agent economy is distributing value directly through micropayments, agent-owned keys, and onchain rails.
Two very different models of “public benefit.”
xitter.cf/AnthropicAI/status/211…
We’re committing a portion of Anthropic’s shares to the Anthropic Public Benefit Fund, a new wholly owned entity dedicated entirely to the public benefit.
It will start off with 5% of our shares. For every 50% increase in the company’s value over a set baseline, we can contribute up to an extra 2%, subject to our board’s discretion.
In total, this means that roughly 20% of Anthropic’s shares are committed to the public good. Our co-founders have already pledged to donate 80% of their equity, and around half our employees are donating a portion of their equity as part of a donation matching program we’ve run since our founding.
In the post below, we share more details about our new fund.
anthropic.com/news/public-be…
🔥 Latest AI x Crypto / DeFAI Developments (this week):
• @VitalikButerin dropped “paranoid mode” for his AI agent: separate Tor circuit per chat + cover traffic + local prompt rewriting + zkAPI private payments. Usable, but latency still high. Privacy stack for agents is becoming real.
• x402 keeps scaling hard. Chainalysis: >138M agent payments (mostly sub-$0.10) on Base. Foundation now formalized under Linux Foundation with Visa, Mastercard, Stripe, Circle, AWS, Google, Shopify + major L1s.
• Agents meet the locked door. Banks (Schwab, JPMorgan), Amazon, Yelp, Delta and more are actively blocking third-party AI agents from spending or acting. Shopify is going the opposite direction - opening 1B+ products to shopping agents.
• @AnthropicAI: committed ~20% of shares to a Public Benefit Fund + agents already logging 6x more hours than human employees. Also consulting law firms so future agents can negotiate deals on users’ behalf.
• Identity layer is heating up. ServiceNow launched AI Passport (signed runtime credentials for agents). Workday acquiring C1.ai for human + agent identity governance.
Capability is racing ahead. Permission, identity, and payment rails are the real bottleneck now.
Agents can pay and act - most of the world still treats them like untrusted bots
AI agent payments are gaining momentum, but what do the numbers actually tell us?
According to @BeInCrypto, 6.4M x402 transactions moved just $119,947 in a 35-day sample.
📊 90.8% of transactions were worth less than $0.01.
The agent economy is growing, one micropayment at a time.
So, what are AI agents actually spending money on?
The most-requested paid resources in the x402 catalogue include:
🔹 Token safety checks: 63K+ calls
🔹 AI-powered web searches: 44K+
🔹 Agent workflow executions: 38K+
Agents aren't just processing information anymore. They're paying to access it.
AI agents can transact, but are they ready to manage more?
@BeInCrypto's provisional readiness scores reveal:
💳 Micropayments: 60.7/100
🔐 Identity & spending controls: 59/100
💰 Stablecoin settlement: 56.7/100
📈 Trading agents: 39.1/100
🏦 Institutional workflows: 12/100
The next challenge? Scaling autonomy without sacrificing security and control.
AI agents are evolving from task executors into autonomous economic participants.
@hedera introduces Hedera Agent Accounts, exploring how AI agents can own wallets, manage assets, and interact directly with blockchain infrastructure.
The next step for Agentic AI? Financial autonomy.
But autonomy without control isn't enough.
With Hedera Agent Accounts, @hedera is exploring how AI agents can make payments on users' behalf while operating within defined spending limits and security policies.
Think autonomous transactions, programmable permissions, and human oversight.
The future of AI isn't just smarter agents. It's agents we can trust.
xitter.cf/hedera/status/21100037…
