🚨 Meta just mogged Gemini 3.8 Flash a few hours after launch Muse Spark 1.3 is already scoring 62 on Artificial Analysis vs 59 for Gemini 3.8 Flash Google finally gets a win for once and Meta immediately pulls this😂

Wow, Meta just dropped Muse Spark 1.3, and Google also dropped Gemini 3.8 Flash today! On the overlapping benchmarks Meta actually looks really strong. Muse Spark 1.3 beats Gemini 3.8 Flash on GDPVal-AA v2 (1754 vs 1545), DeepSWE (75.4 vs 73.7), and OSWorld 2.0 (66.9 vs 59.0). Gemini edges it on Terminal-Bench 2.1 (89.4 vs 88.8). (Google DeepMind) The long-context numbers are probably the craziest part for Meta though: 98.5 on MRCR 256K–512K and 98.1 from 512K–1M, vs 91.5 and 73.8 for GPT-5.6 Sol! I think the craziest one to me is that it beats Opus 5 on DeepSWE and Terminal-Bench, but Opus is a beast on GDPVal and OSWorld. Great Job to the team at meta for cooking on long context!

Top 10 networks by RWA active market cap and market share: 1) @Ethereum : $14.65B (58.4%) 2) @BNBCHAIN: $3.67B (14.6%) 3) @StellarOrg: $2.23B (8.9%) 4) @Solana: $2B (8%) 5) @Aptos: $847.9M (3.4%) 6) @avax: $494.7M (2%) 7) @Arbitrum: $331.5M (1.3%) 8) @Plasma: $303.3M (1.2%) 9) @SeiNetwork: $272.3M (1.1%) 10) @0xPolygon: $270.6M (1.1%) Note: RWA active market cap is calculated as onchain value actively exposed to market risk in the hands of users or protocols.

i can't keep up with this crazy model release cycle anymore actually, you won't either... yesterday: fable 5.1 / mythos 5.1, qwen3.8-max-0902(updated), and the multimodal world model "Atlas" today: gemini 3.8 flash holds SOTA for like 2 hours, then Meta drops Muse Spark 1:3 and tomorrow we might get Astra

🔥 trending coins right now these are the coins currently getting the most attention on the market, according to the trending list: @ponsdotfamily - $0.1227 | +58.4% Dog (Bitcoin) - $0.00125 | +7.4% @cashcat_token - $0.2081 | -4.5% @HyperliquidX - $81.87 | +1.3% @TheIndexFi - $0.01861 | -4.7% @Pumpfun - $0.004697 | -3.9% @Stacks - $0.2802 | +17.0% @solana - $97.04 | -3.8% Pons is leading the list with a +58.4% move over 24h, while Stacks follows at +17% interesting part: being “trending” clearly isn't just about price performance, with several coins on the list trading in the red which one are you watching?

🟥 Things I'm looking at today 🟥 Powered by @yeet Subscribe: Crypto higher, ETH ETF inflow streak ends, Broadcom outlook disappoints Crypto * BTC: 78,893 (+2%) | BTC.D: 59.6% (+0.1%) * ETH: 2,429 (+1%) | BNB: 713 (+4%) | SOL: 102 (+3%) * Fear & Greed: 64 | 24h Liq: $286m * BTC ETFs: +$101M | ETH ETFs: -$48M * Top Gainers: MARSCOIN, BR, AKE, USELESS, PONS * Crypto higher, as ETH ETF inflow streak ends * Strategy holds $66b reserves, second to Berkshire * BTC-Nasdaq correlation falls to 33%: Grayscale * BTC downside limited above $76,350: Bitfinex * Average BTC investor in profit at $80K: Schwab * ETH top pick, expect 3-5x move: Hayes * HYPE enters Hashdex US index ETF at 3.4% weight * Multicoin sells HYPE, has sold 75% from peak * PONS ATH, AI coin winner for on-chain stock pairs * Pons launchpad fees ATH, tops all meme chains * Robinhood Chain DEX volume hits ATH $2.67b * Robinhood Chain RWAs top $100m, 2x in 3 days * Hot money rotating out of AI into crypto: CZ * X foiled mass password-reset attack: US AG * CFTC moves to dismiss CME’s perps lawsuit * Kraken parent delays IPO to Q2 2027 * Wyoming taps Chainlink for FRNT reserves * Tokenised equity holders +134% m/m to ATH * Rain exploit drains $1.1m from card users * Sui protocol Full Sail winds down after exploit * Kalshi seeks CFTC nod for first US WTI perp * NJ asks SC to rule on Kalshi sports contract case * Kalshi faces $500K daily fine in Michigan * DOJ seizes $560K in Hamas funding probe * Singapore proposes ban on stablecoin yield * Coinbase launches 10x BTC futures in Canada Macro * NASDAQ: $26,220 (+0.5%) | Gold: $4,430 (+1%) * Stock market fear at two month high * Trump considers declaring Iran war over * US escorts 40 tankers through Hormuz * US destroyed 28 ships in Hormuz on Tuesday * Rename Hormuz the Trump Strait?: Trump * Iran launches new attacks on US bases * Signs of fresh JPY market intervention * BoJ prefers Quarter-Point rate hike * Global bond rout gathers pace on inflation fears * US Private payrolls data weaker than expected * US subprime auto delinquencies rate hits ATH * Gold, Silver now legal tender in Texas * US gets control of over 65b bbl of Venezuela’s oil * Xi to visit White House on Sep 24: Trump * PBOC fixes CNY at strongest since Feb 2023 * BRICS overtakes G7 in economic output * US moves to isolate Nicaragua * Germany nears destroying all ties with us: Putin * Qatar, UAE turn to LNG ship transfers * 10 years jail for NGOs aiding migrants: Greece * DOJ widens beef price probe to 8 retailers AI & Tech * Broadcom slumps, on weaker Q3 outlook * Researchers fear safety disaster ahead of Astra * OpenAI building automated shutdown for rogue AI * 78% odds OpenAI ships Astra today: Polymarket * Trump admin backs OpenAI in NYT copyright suit * OpenAI sued over Tumbler Ridge mass shooting * Altman warns of AI bubble * Google ships Gemini 3.8 Flash, it 'works harder' * Google avoids break-up in ad-tech antitrust case * Anthropic tightens testing, training safeguards * NYSE used Anthropic Glasswing to fix cyber flaws * Claude Commerce Agents goes open source * Meta drops Muse Spark 1.3, claims beats GPT-5.6 * AI token use will not count in reviews: Meta * Microsoft names half its org 'Agents and Infra' * Wonderful raises $550m at $5b, 2x+ in 6 months * AfterQuery becomes YC's fastest-ever unicorn * AI firms doing 'terrible job' on PR: Bessent * NYC bans AI for students until high school * Snowflake surges after market on stellar Q2 * Nvidia now ~8% of S&P 500, tops 4 sectors * Uber is laying off 10% of staff * Uber to launch robotaxis in London @YEET: Crypto's Casino Just passed $3.2B in lifetime volume! gm

A TON OF THINGS HAPPENED IN THE STOCK MARKET TODAY. Here's a full recap: 1. Broadcom $AVGO said it has a “high degree of confidence” it will ship $350B of AI semiconductors over the next 2 years, even if all 30 GW of demand is not deployed because of land, power, and data center constraints. The company laid out a massive AI growth path, targeting AI semiconductor revenue of roughly $58B in FY26, $115B in FY27, and $230B in FY28 — a 4x increase in just 2 years. Broadcom said every 1 GW deployed by OpenAI or Anthropic could support about $30B in annual AI revenue, while AI networking revenue is expected to grow just as fast as its XPU business. In Q3’26, Broadcom reported revenue of $29.6B, beating estimates of $29.36B and up 86% YoY, with adjusted EPS of $3.32 versus $3.24 expected. AI semiconductor revenue surged 221% YoY to $16.7B, and Broadcom expects that to accelerate to $21.7B in Q4, up 236% YoY, while also targeting $30+ EPS in FY28. 2. Meta $META rolled out Muse Spark 1.3, saying the latest model delivers its biggest coding and agentic performance jump yet. Meta says Muse Spark 1.3 matches GPT-5.6 Sol on Terminal-Bench 2.1 with a score of 88.8 and reaches 75.4 on DeepSWE. The model is available now through Muse Code and the API. Zuckerberg also teased Watermelon and open weights coming next, signaling Meta is continuing to push aggressively on frontier coding agents and open AI distribution. 3. Nvidia-backed neocloud Nscale is telling prospective IPO investors it has roughly $103B in contracted revenue, up from $51B before its $45B compute deal with Anthropic, according to The Information. The contracts average 5.7 years, implying about $18B of future contracted revenue per year, though the figure is not current revenue or formal guidance and was described by one source as “illustrative.” Actual revenue is still much smaller but scaling quickly, rising from about $37M in Q1 2026 to more than $100M in Q2, excluding the Anthropic deal, as Nscale prepares for an IPO that could come as soon as this month. 4. HPE $HPE expanded its Oracle $ORCL AI data center deal, with HPE set to deploy Juniper networking gear across Oracle AI data centers globally under a potential multi-year agreement. The deal covers routing, switching, networking support, and financing for OCI’s AI clusters, regional data centers, and edge networks, while HPE also issued Oracle warrants to purchase HPE shares. Separately, HPE reported Q3’26 revenue of $12.2B, beating estimates of $11.91B and up 34% YoY, with adjusted EPS of $1.11 versus $0.93 expected. Networking revenue surged 74.9% YoY to $2.9B, while Cloud & AI revenue rose 25.4% YoY to $9.0B. HPE guided Q4 revenue to $13.9B–$14.8B, above the $12.96B estimate, raised FY26 guidance, and said AI is becoming a multi-year growth driver as record backlog supports the outlook. 5. Cantor Fitzgerald initiated Tempus $TEM at Overweight with an $80 price target, arguing the company is being mispriced as a traditional life sciences data vendor rather than an AI-powered platform business. The firm says Tempus’ Data & Applications segment has a stronger growth and margin profile than the market is giving it credit for, with 26% growth and 76% gross margins. Cantor believes the segment should be compared more closely to platform peers like $PLTR, $SNOW, $DDOG, and $RDDT rather than slower-growing data-vendor peers, creating a favorable upside risk/reward. 6. Google $GOOGL launched Gemini 3.8 Flash Cyber, a new cybersecurity-focused AI model that scored 86.2% on CyberGym and 47.2% on CWE-Bench. Google says the model has a 70%+ success rate identifying vulnerabilities across 20 programming languages and produced 2.6x more correct patches than larger models on real Chrome security bugs. Initial access is limited to government agencies and cybersecurity partners through Google’s Fairwind program. 7. The top 10 most active options today by contracts traded were $NVDA with 5.7M contracts, $TSLA with 2.8M contracts, $AAPL with 1.7M contracts, $META with 1.0M contracts, $MU with 959K contracts, $AMZN with 732K contracts, $PLTR with 675K contracts, $DELL with 633K contracts, $INTC with 557K contracts, and $GOOGL with 536K contracts. 8. JPMorgan says a 5% 10-year Treasury yield could be the level that starts to pressure stocks, with Grace Peters noting that “5% psychologically has an impact” and could trigger a knee-jerk equity selloff. She sees a potential 5%-8% correction into the midterms, but views that as a healthy pullback rather than a structural break. JPMorgan remains constructive on stocks longer term, with its thesis centered on a capex-driven earnings supercycle, while the key medium-term test for AI will be whether the spending translates into real returns. 9. Berenberg initiated Rocket Lab $RKLB at Buy with an $83 price target, implying about 29%-35% upside, calling it the only end-to-end public pure-play in space across launch, satellites, components, and spectrum. The firm says Rocket Lab has an effective monopoly in dedicated small-lift launch, is entering medium-lift with Neutron, and is benefiting from rapid growth in satellite manufacturing and components as space budgets hit records. Berenberg also said the Iridium acquisition adds scarce global spectrum and recurring applications revenue, while Rocket Lab’s vertical integration gives it long-term optionality not fully reflected in the stock’s valuation. While acknowledging execution risk and a high multiple, the firm called Rocket Lab one of the most compelling long-term assets in the space sector and said it would buy or add on volatility. 10. The 60+ day delinquency rate on U.S. subprime auto loans has climbed to roughly 5.2%, the highest level on record and more than double where it was four years ago. Serious subprime auto delinquencies are now about 1.7 percentage points above their 2008 financial crisis peak, while prime auto loan delinquencies have also risen to around 0.4%, near the highest level since 2011. At the same time, total U.S. auto debt increased by $28B in Q2 2026 to a record $1.71T. 11. Onchain tokenized equity holders have reached a record 1.9M, up 134% month-over-month and 1,360% year-to-date. Just 10 months ago, fewer than 100,000 people held tokenized assets, but demand for 24/7 markets and access to names tied to the record IPO wave, including SpaceX, OpenAI, and Anthropic, has accelerated adoption. Jupiter, the largest onchain trading platform on Solana, has driven much of the growth, with 61% of volume now happening during off-hours and active tokenized equity traders up 46% month-over-month. 12. Microsoft $MSFT will begin disclosing Azure revenue as part of a major FY27 reporting overhaul, shifting from three business segments to two: Agents & Infrastructure and Devices & Consumer. For Q1 FY27, Microsoft expects Azure growth of 44%-45% in constant currency, Agents & Infrastructure revenue of $75.15B-$75.75B, Devices & Consumer revenue of $14.7B-$15.2B, Microsoft 365 Commercial Cloud growth of about 17% in constant currency, and Search & Ads ex-TAC growth in the mid-to-high single digits. The new structure marks Microsoft’s biggest reporting change since adopting its prior three-segment model in FY2016. WALL STREET IS THE GREATEST SHOW ON EARTH.

🧩 Qwen3.8-Flash-Next: A 6B-Active Preview of the Qwen4 Architecture On the same night Zhipu's GLM-5.3-Flash took over the timeline, @Alibaba_Qwen open-sourced Qwen3.8-Flash-Next — explicitly positioned as a preview of the Qwen4 architecture. Zhihu contributor Kitt在进化 argues that for people who actually run models locally, this is the more practical release of the night. The local-deployment groups he is in are, in his words, on fire. His take: the model is called 3.8, but the architecture is really Qwen4 in preview — and it borrows the best ideas from across the field. 1️⃣ Smaller, cheaper, and realistic for local deployment Flash-Next has 125B total parameters with only 6B active — far smaller than the 300B-class GLM-5.3-Flash. The API is priced at ¥1 input, ¥3 output, and ¥0.1 per cached million tokens, roughly matching DeepSeek-V4-Flash's off-peak rates. His comparison, per 1M tokens in RMB — input / output / cache hit: 🔹 Qwen3.8-Flash-Next: 1.0 / 3.0 / 0.1 🔹 GLM-5.3-Flash: 0.4 / 1.4 / 0.115 (limited-time promo rate) 🔹 DeepSeek-V4-Flash (off-peak): 1.5 / 4.5 / 0.05 2️⃣ A Qwen4 preview wearing a Qwen3.8 name The author points out this is the same play Zhipu made: GLM-5.3-Flash's architecture is also completely different from GLM-5.3. Shipping the new architecture as open weights early is deliberate pathfinding. Inference frameworks like vLLM and SGLang, plus the quantization toolchain, all need lead time to adapt before Qwen4 proper arrives. 3️⃣ An architecture that borrows from everyone The author reads Flash-Next as a synthesis of the field's best recent ideas. 🔹 Attention: Qwen's in-house QSA, built to balance throughput and speed on long context. 🔹 Knowledge: it absorbs DeepSeek's Engram line of work, packing prior knowledge into 51B of N-gram side parameters — high knowledge density at minimal compute cost. 🔹 Training: the Muon optimizer, popularized by Kimi, scheduled together with AdamW. 4️⃣ Half the active parameters, still overtaking At 6B active, Flash-Next is nearly half the size of DeepSeek-V4-Flash — yet Qwen's reported benchmarks show it overtaking that model on multiple coding and agent leaderboards. The author says he is not worried about real-world experience: the recent Qwen3.8-27B already proved itself in daily use, and this sits on the same foundation. 🔗 Full Reading: #Qwen# #Alibaba# #Qwen4# #OpenWeights# #LLM# #AIInference# #MoE#