aiagent

🦞 Hermes: Shrimp Farming Memo
🔥 Kimi K3: A new open-source king with 2.8T specs

K3 may be the biggest variable in open-source models this year. For the first time, a Chinese model has taken a seat at the table in global benchmarks of overall intelligence.
In Moonshot AI’s Kimi series, people viewed the K2 era as a competitor to DeepSeek. But with K3, the landscape changes—it directly targets GPT-5.6 and Claude Fable.

Key numbers: 2.8 trillion parameters, 1M token context, multimodal, and full weights opening on July 27.

📊 II. Benchmark performance

Artificial Analysis comprehensive evaluation:
Intelligence Index: 57.1
Coding Index: 76.2
Agentic Index: 50.1

What do 57 points mean? In the same tier as GPT-5.6 Sol (58.9) and Claude Opus 4.8 (55.7), only behind Fable 5 (59.9).

Agent capabilities are even more outrageous:
- GDPval v2 → Elo 1668, surpassing GPT-5.5 and Opus 4.8
- AutomationBench-AA → #1 worldwide
- AA-Briefcase long-task knowledge work → #2 worldwide
- Frontend Code Arena → #1 worldwide (jumping from #18 in K2.6 to #1)

Last year it was about who could chat better; this year it’s about who can get work done better.

💰 III. Price comparison

Kimi K3 (official direct) $1.10 input $5.50 output AA 57.1
Kimi K3 (OR) $3.00 $15.00
DS V4 Pro $0.44 $0.87 44.3
DS V4 Flash $0.14 $0.28 40.3
GPT-5.6 Sol $5.00 $30.00 58.9
Claude Opus 4.8 $5.00 $25.00 55.7
GLM-5.2 $0.94 $2.96 51.1

- 8~20x more expensive than V4 Flash, but Index is higher by 17 points
- 4~5x cheaper than GPT-5.6 Sol, only lower by 1.8 points
- 5~7x more expensive than K2—the official no longer sells it as a “cheap big bowl”

🔓 IV. The significance of opening weights on July 27

The impact comes in three layers:
① Individual developers — A 2.8T parameter model can’t be run locally. You’d need an 18~70 H100 cluster; it’s not feasible for individuals.
② Cloud providers / API middlemen — With the weights, they can build their own inference services, drastically reducing access costs, making price competition possible.
③ Open-source ecosystem — If it’s delivered, K3 will become the open-source model with the strongest overall capabilities, surpassing all open-source contenders like GLM and DeepSeek.

💡 V. Another overlooked piece of data

Compared with K2.6, K3 uses about 21% fewer output tokens, yet achieves higher intelligence scores. This isn’t just “more parameters”—there are clear improvements in inference efficiency and training methods as well.

⚠️ VI. Uncertainty

- Will the weights be delivered as scheduled on 7/27?
- The open-source license terms for a 2.8T model? Commercial restrictions?
- Under the MoE architecture, what is the real-world inference efficiency?
- After API prices increase, will it be able to retain users?

🔮 Summary

K3 isn’t necessarily already the strongest model in the world, but it very likely is a top contender for the world’s strongest open-source model. If the weights are opened on time, the gap between open-source and closed-source will shrink further in the second half of the year. The fault line for Agent capabilities is here—this year’s competition is about who can get work done, not whose parameters are bigger.

#KimiK3 #开源AI #AIAgent

The trading volume of recent on-chain AI agent projects has jumped 3x month-over-month, but after going through dozens of contract codebases, I found an awkward truth: in most agents, the core trading function has no authorization/permission layer at all. The agent can call any contract it wants and transfer any token it wants. This so-called “autonomy,” which sounds impressive, is actually just a bot running around completely exposed—with a private key.

When something truly goes wrong, who will stand behind it? Last month, one agent project lost control for a prompt injection: within 5 minutes, it swapped the treasury’s stablecoins into some worthless meme token, and then it never came back. Even if the post-incident audit is done beautifully, the money is already gone. For institutional strategies, this kind of black swan is terrifying—it can wipe out months of alpha in one shot.

I discussed this with the product team at @NewtonProtocol, and they’re doing something I think is on the right track: turning the authorization layer into a hard gate before settlement. Before each on-chain transaction is finalized, the AI agent must pass a Newton attestation check. It verifies whitelisted assets, maximum drawdown, approved counterparties—if any requirement isn’t met, the signature is rejected immediately. This isn’t just after-the-fact alerting; it blocks the transfer before any real money moves, preserving funds in black and white. Institutional-grade RWA, stablecoin compliance flows, and AI autonomous strategies can all be built on this framework.

I looked at their Mainnet Beta on the TokenizeThis NYC conference in June. With VaultKit, this SDK provides vault creators with a complete Rego policy template: whitelisting, maximum drawdown (X%), approved counterparties, and other rules. Just one line and the rules are active. Combined with attestation, every agent rebalancing transaction must pass the policy check before it can be committed on-chain—violations are rejected outright. With AI autonomy plus these on-chain hard rules, the two finally can be discussed together, and cross-chain strategy consistency can also be ensured using the same set of Rego rules.

Right now, $NEWT is hovering around $14.82, up 4.27% over the past 24 hours. Even after more than a month since mainnet launch, liquidity is still steadily climbing. The token itself handles the staking, gas, and governance “three-piece set,” and the tokens from the 2025 Binance HODLer airdrop are also starting to become active in the secondary market. For project teams building agents, integrating VaultKit into the policy layer may be more meaningful than obsessing over the two words “autonomy”—because code doesn’t lie, and contracts don’t do sentiment.

#Newt #AIagent #NewtonProtocol #链上授权 #DeFi

The Talus ecosystem has clearly gained momentum recently. After the ACC accelerator program was completed, the $US token airdrop and vesting were implemented in sync, and the basic short-term price catalysts are basically all in place.

Let’s break it down into a few points:
First, the pace of ecosystem development is faster than the market expected, so there’s actually something to talk about narratively;
Second, after the airdrop was released, selling pressure actually eased, suggesting the token holders’ structure isn’t as fragile as people imagined;
Third, overall market sentiment has warmed up, and the AI Agent track is being picked up again.

Current price is $0.02346, market cap is $51.61 million, and 24h trading volume is $9.32 million. Liquidity looks healthy for this size, and there is some room for short-term positioning, but you need to keep an eye on the vesting release schedule—don’t treat the catalysts as a long-term thesis.

The AI + Agent track is still filtering projects. This Talus move is arguably a showcase of execution.

#Talus #AIAgent

Circle CEO Jeremy Allaire’s 《Agency Economics》 is, in essence, drawing a three-layer architecture diagram for the future economic system.

The bottom layer is the currency layer—stablecoins take over bookkeeping and settlement, bypassing the national central bank. The middle layer is the economic operating system—smart contracts move “execution” from courts into globally runnable code. Only at the top is the AI Agent, responsible for replacing local labor with borderless computing power.

What I care about most is the focus within this narrative. The market is used to treating AI and crypto as two separate tracks to trade, but Allaire’s judgment is very straightforward: AI only provides “mechanized intelligence.” Without the accompanying on-chain property rights, payments, and credit infrastructure, an Agent will always be just a hired script—it can’t become an independent economic entity.

Conversely, this also explains why Circle is betting on $USDC as the “accounting unit for agency economics”—stablecoins are not the end point. They are the only money the AI Agent can use when it goes to work.

Once these three layers truly run end to end, it’s not just payments that get restructured, but banks, courts, and the labor market all replaced by software at the same time. That’s the longest snow slope in the stablecoin track.

#Circle #稳定币 #AIAgent

Q402 roadmap is taking Quack AI from concept to reality: Gasless payments are live to lower the usage barrier, the Agentic Wallet upgrade makes intelligent agents easier to use, and the DAO governance framework brought by ERC-8004 also makes community power more clearly defined.

More importantly, Robinhood Chain’s expansion and the $Q token, which directly receives a 50% subscription discount in payment scenarios—this kind of "real consumption" is the most lacking part of today’s AI narrative.

Current price is $0.01976, market cap is $82.39M, and 24h trading volume is 5.63M. The market cap-to-volume ratio is healthy, indicating there is genuine turnover rather than market-making support alone. If these lines in Q402 can be delivered on schedule, the adoption curve for the ecosystem will likely show up before the price does, making it worth putting on the watchlist for the AI Agent track.

#QuackAI #AIAgent #Web3

EVAA Protocol Recent performance can be summed up in one word: tearing apart.

The concept of AI Agents has brought in short-term capital in waves, pushing $EVAA into violently two-way fluctuations. The current price is $2.47. Over the past 24 hours, trading volume surged to $82.43 million, while the market cap is only $44.18 million—trading volume is already nearly double the market cap. This is a classic high-turnover speculative structure.

With both pump and dump coexisting, the core reason is that the contract-side position valuation is clearly inflated. Once the price deviates, it triggers a chain reaction of liquidations, producing a pin-trading pattern in a short period of time. That’s also why community sentiment is extremely polarized: some see it as the next breakout point for the AI Agent track, while others question whether the fundamentals can actually support such a valuation.

In the short term, rather than trying to guess the direction, it’s better to watch two things: first, whether the contract funding rate and open interest continue to rise; second, whether spot trading continues to expand. If the former gets overheated, it means risk is accumulating—only the latter reflects real follow-through capacity.

In high-volatility products, position size always matters more than direction.

#EVAA #AIAgent #high volatility

$EVAA In recent trading, the market has shown typical high-volatility characteristics. Sudden surges and crashes in the short term have caught many people off guard.

Current price is $2.47. The 24h trading volume has jumped to $82.43 million, while the market cap is only $44.18 million—an unusually inflated turnover rate. This, by itself, is a signal of short-term capital speculation.

Three observations:

1. The AI Agent narrative has reignited again, drawing short-term funds in a rush and pushing price sentiment into an overheated zone;
2. On the derivatives side, open interest is clearly overstretched versus spot support. If the direction reverses, liquidation cascades are triggered one after another, and the pullback magnitude is amplified by leverage;
3. Community sentiment is highly polarized. Longs firmly believe the narrative will continue, while shorts are betting on valuation mean reversion. With no middle-ground consensus, price action naturally swings violently.

For me, this kind of asset is not suitable to chase. It’s better to wait for volatility to contract and for trading activity to cool back into a normal range before reassessing the structure. What’s truly worth paying attention to isn’t individual candles, but whether the ratio of market cap to trading volume returns to rational levels.

High volatility isn’t an opportunity by itself—position sizing is what matters.

#EVAA #AIAgent

$EVAA The recent market action has been quite chaotic. A wave of short-term capital has poured into the AI Agent concept, rapidly pushing the price up. However, the open interest on the derivatives side is clearly overheated. Once it pulls back, it triggers a chain reaction of liquidations—bidirectional price spikes have become the norm.

Current price is $2.47, with $82.43 million in 24-hour trading volume. The market cap is only $44.18 million. Trading volume is nearly double the market cap—this structure alone indicates that the chips are churning at high frequency, with sentiment driving the move far more than fundamentals.

Community disagreements are also obvious: some people believe the AI Agent narrative can still keep things going, while others worry that valuations are being overspent. From a personal perspective, at this stage position sizing matters more than direction. Chasing longs higher can easily be反噬 by liquidations, while aggressive short positions also need to guard against a second ignition of the narrative. Wait until volatility converges and open interest data returns to rationality before acting—then the win rate will be much higher.

#EVAA #AIAgent #DeFi

$EVAA Recent price action has been nothing short of a roller coaster—both sides got swept with back-and-forth needle moves, harvesting both bulls and bears.

Break it down and the logic isn’t complicated:
First, AI Agent narratives have reignited, and short-term capital rushes in, quickly pushing prices higher;
Second, positions on the derivatives side are clearly overheated—once leverage is piled up, any small gust triggers a chain liquidation, making it fall faster than it rose;
Third, community sentiment is sharply polarized: the bulls see it as the start of a new track, while the bears think it’s the operator distributing—without a clear mid-term judgment from anyone, volatility gets amplified further.

Current data: price is $2.47, 24h trading volume is $82.43 million, and market cap is only $44.18 million. Trading volume is already nearly twice the market cap, indicating extremely fast turnover of chips—classic high-emotion, low-support structure.

The given trading guidance is quite straightforward—
· If you want to participate, use a small position size and a wide stop-loss; don’t open high leverage to gamble on direction;
· For spot players, it may be better to wait for trading volume to cool off and for the candlesticks to converge before assessing whether the trend is actually valid;
· Watch the overall sentiment of the AI Agent sector—once the sector leader goes quiet, pullbacks in follower tokens are likely to be even harsher.

Small market cap + strong narrative + high leverage have always been a double-edged sword—the timing matters more than the direction.

#EVAA #AIAgent

Open the gainers leaderboard: the top 10 are packed around a token doing “#AIAgent ,” namely $VIRTUAL . It’s up 14.3% today, and the current price is $0.62.

This kind of move looks a bit strange against today’s broader market backdrop. BTC is only down -0.15% today—basically flat. FET (another AI token) is down 0.3%, NEAR is down 0.5%, and TAO is also down 0.3%. The AI agent track isn’t seeing an overall rise today; instead, only $VIRTUAL is clearly moving, and the volume is also pretty sizable, with $14.78 million in trading value.

When a single token is the only one rallying within a niche like this, I feel both cautious and intrigued at the same time.

$VIRTUAL is Virtual Protocol’s token, running on the Base chain. Its main focus is enabling people to create, deploy, and trade AI agents. During last year’s Q4 hype cycle for AI agents, this thing surged to several dollars. Now it’s at $0.62, far below its peak—basically in a “sitting near the lows” state.

Recently, on Twitter, people have started discussing AI agent infrastructure again. Most of the conversation is about which chains can truly support agent operations and which ones are only “riding the narrative.” Once this kind of discussion heats up, the related tokens often move before the real-world data catches up—an old rule of crypto. Whether it’s genuinely useful is another question, but as a signal, that phenomenon is objectively real.

Today, $VIRTUAL ’s 24h high reached $0.6485, and it’s now hovering around $0.62. People who chased the price today are currently sitting on a loss and haven’t managed to reclaim the level. From this price structure, the bulls may have intentions, but they haven’t fully taken control yet.

My personal view is: if the AI agent track truly regains momentum in the second half of the year, $VIRTUAL is likely to be one of the leading movers—because in the last wave it was the first to start running. However, this view is predicated on the narrative continuing to build; you can’t infer much from just one day of data.

Next, it’s the changes in trading volume over the next 3–5 days that are more worth watching than how much it’s up today. If volume keeps up, then the narrative can be said to have truly returned.

VIRTUAL—This breakout surge has logic behind it that’s more than just short-term sentiment.

Robinhood has announced that it will soon support “proxy trading.” Eligible U.S. users will soon be able to connect an AI Agent to a dedicated account so it can place orders on their behalf, while still retaining real-time profit/loss and push notifications. In other words, it’s bringing AI Agents to the retail trading front lines for real.

$VIRTUAL , as a representative in the AI Agent sector, is giving a direct reaction on the chart: it broke above 0.63 USDT and is currently at 0.6322, up 18.83% over the past 24 hours.

My observations:
- This round isn’t just pure meme sentiment—it’s a narrative mapping of “AI + a brokerage compliance entry point”
- Robinhood’s move suggests that AI Agent trading is starting to gain a mainstream compliance channel, rather than remaining only an on-chain experiment
- In the near term, watch for profit-taking. Above 0.63 is a dense trapped-position zone; whether it can hold above that will determine the space for the next leg

The narrative is moving, the chips are changing hands—don’t just stare at the candlesticks. Figure out clearly who is handing this sector a microphone.

#VIRTUAL #AIAgent #Robinhood

VIRTUAL quietly pulled up a big bullish candle, directly breaking above 0.63 USDT—up 18.83% in 24 hours—and the current price is 0.6322.

This move isn’t random—Robinhood has just officially announced bringing crypto into the "proxy trading" era. Eligible US users will soon be able to connect their AI Agent to a dedicated Robinhood account, letting it handle buying and selling for you, with P/L notifications and real-time tracking, leaving nothing out.

AI Agent + on-chain assets—once this narrative is formally adopted by mainstream brokers, AI-agent infrastructure projects like Virtuals Protocol will be right in the middle of the headwinds and the boom. The market clearly smells it, and the funding has moved very directly.

What to watch is: above 0.63 is a previously tightly trapped consolidation zone. Whether it can hold steady matters more than how much it pumps. If Robinhood later discloses more details, there may be another round of fuel for sentiment; otherwise, short-term profit-taking could also be significant.

The AI Agent track is moving from concept to real trading use cases, and this year may only be just getting started.

$VIRTUAL

#VIRTUAL #Robinhood #AIAgent

Metis’s recent performance is interesting—two lines are moving at the same time.

First is the ecosystem side: an upgrade to AI Agents has given on-chain storytelling new traction, with both developer activity and interaction scenarios being repriced; second is its subsidiary, Metis TechBio, which secured a $1.6 billion drug licensing agreement, turning the “blockchain + biotech” combination from an idea into an expected cash-flow outcome.

Currently $METIS is trading at $2.73, with $2.15M in 24h volume, and a market cap of only $20.61M—it's not a big float. Once market recognition catches up with the fundamentals, upside could be fairly noticeable.

The key points to watch are whether the on-chain data enabled by the AI Agent upgrade can be sustained, and how quickly TechBio can commercialize and roll it out. A ticker driven by both narrative and fundamentals—relatively scarce in this round.

#Metis #AIAgent #RWA

Talus’s ecosystem is quietly gaining momentum. The first accelerator, Talus/acc, together with the Sui Foundation and Cicada Capital, selected seven teams from a large pool of projects, focusing on AI Agents, RWA, and on-chain data. This isn’t just a simple incubation showcase—it’s about turning technical narratives into verifiable, real-world capabilities.

Based on the data: $US is currently priced at $0.02765, with a 24-hour trading volume of $9.36 million and a market cap of $60.84 million. The trend is steady-to-upward. While the market cap isn’t large, liquidity appears relatively active, suggesting that the market continues to pay attention to its AI Agent direction.

Personal view: The quality of the accelerator determines the ecosystem’s depth. If among these seven projects, one or two genuinely deliver AI Agent applications with real user retention, Talus’s technical trustworthiness could be re-priced. Conversely, if it’s only a published list and no lasting traction, the hype will dissipate quickly. The project disclosure cadence over the next one or two months is a key point to watch.

I’m not chasing the top, but it’s worth adding to the watch list.

#Talus #AIAgent

Talus生态又有新动作。首个加速器Talus/acc联合Sui基金会和Cicada Capital，筛选出7个聚焦AI Agent、RWA和链上数据整合的项目，直接补强了生态落地能力和技术叙事的可信度。

目前 $US 价格0.02765美元，市值约6084万，24小时成交量936万，盘面稳中有升。加速器落地往往是叙事催化的起点，接下来要看这7个项目能不能真正跑出成果，把生态含金量抬上去。

AI Agent赛道竞争激烈，Talus选择用加速器绑定优质团队而不是单打独斗，这个路径比自建产品更有杠杆效应。短期看情绪面，中期看首批项目的兑现节奏。

#Talus #AIAgent #Sui

Prime Intellect raises $130M in a Series A funding round

#PrimeIntellect secures $130M in a Series A funding round led by Radical Ventures, with participation from NVIDIA Ventures, Intel Capital, and Dell Technologies Capital. The funding round brings Prime Intellect's total funds raised to $150.5M.

Prime Intellect is a decentralized #AI infrastructure platform building an open superintelligence stack with tools for #AIagent development, compute, reinforcement learning, and deployment.

👉 x.com/PrimeIntellect/status/2074899489190785419

OpenAI throws out its trump card: the GPT-5.6 trio makes a group appearance

The flagship Sol, the balanced Terra, and the low-cost Luna have all rolled out on ChatGPT, Codex, and the API. Pricing follows a tiered ladder—what you spend is what you get.

Most ruthless is Sol’s Ultra mode—by default it schedules four AI agents to work in parallel: writing code, doing research, running security audits, and handling knowledge-based tasks, all delivered in one go. The era of single-agent systems is basically over.

The inspiration for the crypto industry is very direct:
- On-chain analysis, contract audits, and strategy backtesting—multi-agent collaboration will become standard
- The AI + Crypto race will drive computational power and reasoning demands to the next level
- The barrier for developers to build agents with the API drops another notch

Model capabilities jump every half-year, and the window for project teams to tell stories keeps getting shorter—only those that can produce data will command a premium.

What would you have Sol help you with first? Let’s chat in the comments.

#OpenAI #GPT5 #AIAgent