Frontier Models Ignite Price War | Rewire Morning News Brief

Anthropic and OpenAI Cut Prices on the Same Day, AI Competition Shifts from Capability to Cost. Trump Renames AI "Superintelligence" at the UN General Assembly, Rejects All International Regulatory Frameworks.

1 | Anthropic and OpenAI Release Price-Cut Models on the Same Day, AI Competition Officially Enters a Price War

Anthropic and OpenAI released new price-cut models on the same day. Opus 5.5's API pricing dropped to $4 per million tokens for input and $20 for output, 20% lower than Opus 5, with cache read costs slashed by 60%. Anthropic says it approaches Fable 5.1's performance on most tasks, with total running costs 40% lower, a 1 million token context window, and output speed improved by over 30%.

OpenAI launched GPT-6 Sol and Luna the same day. Sol is priced at $2/$10 (per million tokens input/output), while Luna is more aggressive at $0.1/$0.5, available to free users. Sol's error rate is halved compared to its predecessor, and Luna targets high-throughput scenarios. Both models inherit Astra's training methods, with clearly defined roles.

Ars Technica's headline precisely captures the logic of this round of competition: "A little more capability, a lot less money." Frontier capability is no longer the bottleneck—deployment scale is. Opus 5.5 doesn't sell "stronger," it sells "close to the strongest but much cheaper." Luna pushes free users to the level of paid users from a year ago. The price cuts are about capturing scale effects on the inference side.

(Sources: Anthropic / OpenAI / Ars Technica / Fortune / VentureBeat)

2 | Trump Rejects Global AI Agreement at the UN General Assembly, Announces Renaming AI "Superintelligence"

In his speech at the United Nations General Assembly, Trump rejected all international AI governance agreements, calling them a "globalist conspiracy." He announced that the United States would rename "artificial intelligence" to "superintelligence," reasoning that "artificial sounds fake, and it's not fake." He said whoever masters superintelligence is the winner, and the United States is far ahead of China and all other countries.

UN Secretary-General Guterres, the UK, France, Canada, and even several leading AI labs have all called for establishing an international regulatory framework. Trump's response was that existing U.S. institutions are sufficient to handle any risks, and no new agreements are needed. He compared AI to the Industrial Revolution, stating his position is to "encourage rather than restrict." Scientific American noted that his stance on AI is almost completely identical to his stance on climate change.

The remarks came less than a week after the U.S. and China agreed to establish an "AI incident hotline." The U.S. rejects multilateral frameworks on one hand while needing bilateral mechanisms with China on the other. Those who reject rules don't want no rules — they want to set the rules themselves.

(Sources: Fox News / Scientific American / NBC News / Bloomberg)

3 | Alibaba unveils "China's most powerful" AI chip Zhenwu V900, targeting trillion-parameter models and 20 gigawatts of compute

Alibaba has unveiled its next-generation AI chip, the Zhenwu V900, claiming it is "China's most powerful AI chip." It delivers 3x the performance of its predecessor and supports cluster training of up to 500,000 chips. Alibaba simultaneously announced plans to train a 10-trillion-parameter AI model, with executives explicitly stating at the launch event their pursuit of artificial general superintelligence. The company aims to build over 20 gigawatts of global data center capacity by 2032.

The timing is noteworthy. The Trump-Xi meeting at the UN General Assembly is set for this week, and AI chip capability is precisely the core issue in the two nations' rivalry. Last year, Huawei shelved its Ascend global rollout because domestic capacity was maxed out. Alibaba's entry means China no longer has just one company capable of making high-end AI chips.

From being blockaded to achieving self-sufficiency, the narrative around China's AI chips is shifting phases. Alibaba didn't start as a chip company — it's an internet platform. When platform companies start making their own chips, it mirrors the path taken by Apple and Google, with the difference being that Alibaba is doing this under the premise of having its supply cut off.

(Sources: Nikkei Asia / South China Morning Post / Fortune / Bloomberg)

4 | SoFi becomes first licensed bank to use stablecoins for card settlement; Binance invests $100 million in Circle

SoFi Technologies has become the first licensed U.S. bank to use its own stablecoin for bank card settlement. SoFi uses its own dollar stablecoin, SoFiUSD, to settle debit and credit card transactions on Mastercard's global payment network, with annualized transaction volume exceeding $25 billion. SoFiUSD launched last December and is the first stablecoin issued directly by a consumer banking app in the U.S. SoFi holds a national bank charter from the Office of the Comptroller of the Currency.

On the same day, Binance made a $100 million strategic equity investment in stablecoin issuer Circle. Circle is the issuer of USDC, and Binance is the world's largest crypto exchange. The two previously had public friction over USDC's delisting on the Binance platform, and the $100 million investment signals that friction has given way to mutual interest.

The inflection point for stablecoins isn't the coin price — it's that traditional financial institutions are starting to use them as infrastructure. SoFi isn't a crypto company; it's a consumer finance platform with a national bank charter. By choosing blockchain settlement over traditional clearinghouses, the stablecoin narrative shifts from "crypto goes mainstream" to "mainstream finance is already using crypto rails."

(Sources: Finextra / Fortune / PYMNTS / The Block)

5 | Iran Signals De-escalation at UNGA, Oil Drops Below $100, Trump Gives "Post-Election" Timeline

Brent crude fell to $101.75 and WTI dropped to $98.34, both down about 2% on the day. Markets are betting on diplomatic progress between Iran and the U.S. during the UN General Assembly. Iran's delegation attended multiple UNGA events, and the two sides made contact.

But Trump simultaneously threatened Iran in his UNGA speech. He told world leaders, "I believe we'll make a deal after the election, because they have no reason not to." What the market heard wasn't the threat — it was the words "after the election." This sets a countdown for all parties involved: no substantive breakthrough before November 3, but the negotiating window after that is already being priced in ahead of time.

Saudi Aramco's pipeline was recently attacked, some European customers were notified of supply cuts next month, and Saudi Arabia was forced to increase exports through the Strait of Hormuz. Pipelines bombed, dependence on the strait rising, yet oil prices falling — the market's pricing logic has already shifted from "supply disruption risk" to "the probability that the war can't be sustained."

(Sources: Al Jazeera / NBC News / Reuters / Offshore Technology)

Also Worth Knowing ↓

Altman and Amodei will brief the UN Security Council on AI safety on Wednesday. France, as rotating president, organized the session, covering risks of AI going out of control and being misused. The same week, both men's companies released price-cut models — the people building products and the people making rules are increasingly the same crowd. (Sources: Quartz / CNBC)

a16z launched the Horowitz Andreessen Academy, with Anthropic, OpenAI, Google, Meta, Nvidia and 10 other companies as founding partners. $35 million in funding, each student receives over $50,000 in compute subsidies and a San Francisco residency slot, with no degree awarded. VC is shifting from investing in companies to directly cultivating company founders. (Sources: TechCrunch / a16z)

Toyota plans to deploy 400,000 humanoid robots in its factories starting in 2028, with an annual investment of $6.4 billion. Workers are being asked to train these robots to perform their own daily tasks. Toyota says the goal is "collaboration," not replacement. (Source: Ars Technica)

Taiwan region's August export orders hit a record high, with AI chip demand as the main driver. TSMC's share of the region's exports continues to climb, further deepening the geopolitical concentration of the semiconductor supply chain. (Source: Reuters)

The Nasdaq hit a record high. Against the triple backdrop of AI cost-reduction signals, falling oil prices, and the Fed holding rates steady, risk appetite for tech stocks is recovering. (Source: Reuters)

The Meta Muse vulnerability has been patched, with users surpassing 500,000. (Continued from yesterday's report) After being breached two weeks post-launch, Meta rolled out an emergency patch and added a human concierge approval layer. Security researchers note that the root cause of the Secure VM bypass has not been fully resolved. (Source: The Verge / TechCrunch)

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia