奥特曼「红色警戒」5个月后，GPT Image 2屠榜，断层领先反杀谷歌

marsbitPublicado em 2026-04-27Última atualização em 2026-04-27

【导读】被Google按了半年头,OpenAI终于祭出一记反杀。GPT Image 2上线12小时,就登顶Arena文生图榜,领先Nano Banana 2达241分。Arena官方称,这是Image Arena文生图排行榜迄今最大的分差。

发布当天,三榜通杀。

GPT Image 2上线12小时,Text-to-Image(文生图)、Single-Image Edit(单图编辑)、Multi-Image Edit(多图编辑)三个分榜全部登顶。

Arena官方原话:「a clean sweep」(全胜)。

文生图主榜,GPT Image 2 1512分,Nano Banana 2 1271分。241分差距,Arena史上最大。

「没有任何模型曾以这种差距统治过Image Arena」,Arena官方表示。

在Image Arena所有盲测对决中,GPT Image 2的胜率是93%:100张图配对盲选,93张人们选了OpenAI那张。

「如果把DALL-E看作洞穴壁画,把Images 1.0视为古代艺术,那么Images 2.0就是文艺复兴」。

OpenAI在发布会开场中这样介绍Images 2.0,奥特曼更是将它称作跨代升级:

这好像一下子从GPT-3跃升到了GPT-5。

https://www.youtube.com/watch?v=sWkGomJ3TLI

OpenAI官方API文档对Images 2.0给出了一个最高级的评价。

https://developers.openai.com/api/docs/models/gpt-image-2

但真正的故事,并不在数据里。

被Google压了半年,OpenAI总算扳回一局

时间倒回2025年8月。

Google放出了Nano Banana。这个在Gemini里嵌入的图像生成模型,在C端瞬间引爆。

三个月后的Q3财报会上,Google CEO Sundar Pichai亲口披露了一组数字:Gemini月活,从7月的4.5亿涨到10月的6.5亿。

Google Labs负责人Josh Woodward称,这一增长很大程度上来自Nano Banana带动的图像生成热潮。

11月,Google再发Nano Banana Pro。文本渲染能力惊艳,AI图像第一次能把字写对,OpenAI在C端被反超。

11月18日,Google再补一刀。Gemini 3发布即登顶LM Arena,1501分,首个突破1500的前沿模型。

这一月底,奥特曼对全公司发了一份「红色警戒」(code red)的内部备忘录。

据The Information报道,奥特曼私下告诉员工,Gemini 3可能给OpenAI带来经济逆风。Yahoo Finance后续披露:code red之下,OpenAI暂停了AI Agent等其他产品的研发,资源全部倾斜到ChatGPT。

12月,OpenAI仓促拿出GPT Image 1.5。Arena第一,但C端没能引爆。

2026年2月,Google再补一刀,Nano Banana 2登场,Arena再度领先。

OpenAI又输了一次。

一直到4月21日,GPT Image 2上线,OpenAI这才实现反超,重新扳回一局。

画图AI将被重新定义

GPT Image 2凭什么能领先241分?

核心答案藏在架构层面。

GPT Image 2不是Stable Diffusion那一代的扩散模型。

OpenAI研究负责人Boyuan Chen称这是「revamped from scratch」(从零重构)的「generalist model」(通用模型),OpenAI的内部叫法是「图像版的GPT」。

但Chen在press briefing时拒绝公开承认它具体是扩散还是自回归架构。

外界普遍把它理解为「带推理规划的图像生成系统」:画之前先规划,再下笔。这正是GPT Image 2和上一代图像模型最大的不同。

OpenAI在官方说明里给了它一个新标签:首个具备原生思考能力的图像模型(image model with native thinking capabilities)。

画之前先想、画完自己检查、需要时联网搜索资料、一次能产出8张前后连贯的图。

这不是画笔,是会思考的视觉助理。

Arena榜单分项数据显示:

文字渲染(Text Rendering)单项,GPT Image 2比前代涨了316分;卡通动漫和人像各涨296分;3个产品/3D/写实分类,整体在+247到+277分区间。

文字渲染是2025年11月Nano Banana Pro首次解决的问题,但当时准确率94%。GPT Image 2把它推到了99%。

OpenAI发布会现场演示:让GPT Image 2画一碗米饭,其中只有一粒米上写有模型名字。

具体到能力展示,OpenAI总裁Greg Brockman在自己的X账号上做了示范。

第一个案例,老照片修复。

褪色发黄的家庭老照片,一个提示词,立刻变身高清彩色版。

OpenAI官方API文档里那句「high-fidelity image inputs」(高保真图像输入),说的就是模型对原图细节的保留能力:输入端能精确读取褪色的、破损的、模糊的老照片细节,输出端才能重新渲染出清晰版。

第二个案例中,Brockman转发了用户@doodlestein的一组测试图:用同一个复杂提示词让GPT Image 2画一张数学解释图。

他评价说,即便是复杂提示词,GPT Image 2也能生成风格各异的图。

@doodlestein 测试GPT Image 2用同一个提示词画一张线性代数解释图。模型一口气画出4个完全不同的版本:同样是Mona Lisa+特征向量教学,每个版本的构图、配色、信息密度完全不同。

这个案例真正价值不在「能画数学图」,而是解决了过去两年中AI生图的一个重要的痛点:输出单一、变体可控性差。

GPT Image 2第一次让「一个prompt给我4个完全不同的方向」变成了产品级能力。

业内一位LM Arena资深测试者点评道:

GPT Image 2和Nano Banana Pro之间的差距,跟Nano Banana Pro和DALL-E之间的差距一样大。

跨了整整一代。

GPT Image 2 Thinking模式生成的manga风格漫画页:从一个简单提示词出发,模型保持角色一致性、铺出多格剧情。

DALL-E退役,Adobe Canva被逼到墙角

发布当天,下游工具集成的速度比技术圈预期的还快。

Figma、Canva、Adobe Firefly、fal、Hermes Agent,全部在4月21日当天完成集成。

API定价更是暗藏杀机:

高质量出图$0.21一张;ChatGPT Plus $20一个月,图像生成已经包含在套餐里。

这个差价背后,可能带来2026年图像生成行业最大的产业重构。

GPT Image 2生成的photorealistic candid(写实抓拍)。海岸、阴天、复古车、胶片质感——这种过去要专业摄影师外拍+后期才能达到的视觉效果,现在API $0.21一张。OpenAI研究员Gabriel Goh说,photorealism是他对这个模型最兴奋的能力。

5月12日,DALL-E 2和DALL-E 3正式退役。

它们是2022年开启了整个AIGC视觉革命的开山祖师。三年后,被OpenAI自家的继承者,亲手送入历史。

OpenAI在官方发布说明里提到:

图像不是装饰,是语言。一张好图做的事和一个好句子一样:选择、排列、揭示。

这代表了一种产品哲学的转向。

当然,也不是没有反方声音。ZDNet在实测中发现,GPT Image 2无法准确复刻品牌logo,连ZDNet自己的logo都被画歪了。

Nano Banana 2在portrait realism和multi-reference一致性上仍有优势。

GPT Image 2虽然还不够完美,但赛道格局已经出现了变化。

渲染时代结束了,推理时代刚开始

Google把推理塞进图像模型里。OpenAI把图像工具塞进推理模型里。242分Elo差距测的就是二者架构上的差异。

implicator.ai的这句评价,划分了图像生成的两个时代。

2022到2025年,是渲染时代。

DALL-E、Midjourney、Stable Diffusion,目标都是「画得像」。模型是画笔,用户是画师,prompt是画稿。

GPT Image 2代表的是一个推理时代。

模型先思考再下笔,能搜索、能自检、能完成任务。它不是画笔,是会画画的助手。

GPT Image 2发布真正值得重视的,是图像生成走向「会思考」这件事本身。

短期看,Black Forest Labs(Flux 2)麻烦可能最大。

Kingy AI直言:作为diffusion-first(扩散为先)的厂商,Flux 2的整条技术流水线在架构上和「token-by-token」的推理路线是冲突的。

要么融合,要么重写,没有第三条路。

中期看,Google可能会在下个季度反击。Nano Banana 3,或者Imagen-Reason,时间不会太久。

长期看,这件事的影响远不止图像生成。

当AI开始用「思考」来产出图像、视频、音频、代码,整个生成式AI的范式都会跟着发生变化。

去年12月,奥特曼在备忘录里敲下「code red」的时候,应该没想到五个月后会以这种方式回到Arena榜首。

但这次反杀的真正意义,可能不是OpenAI赢了Google,而是OpenAI改写了图像生成赛道的规则。

Arena.AI单图编辑榜(Image Edit Arena):GPT Image 2 (medium) 仍以1510+分继续登顶,第二、三、四、五名全部被OpenAI自家模型和Google Gemini系列占据。https://arena.ai/leaderboard/image-edit

Google下一拳什么时候出?这个问题决定了2026下半年AI格局的走向。

而在那一拳挥出来之前,GPT Image 2会在Arena榜首坐多久,没人知道。

参考资料:

https://x.com/gdb/status/2048449695622586576

https://arena.ai/leaderboard/image-edit

本文来自微信公众号“新智元”,编辑:元宇

Criptomoedas em alta

CitreaCTR

wrapped stUSDTWSTUSDT

Velodrome FinanceVELODROME

BrevisBREV

ZRX（0X）ZRX

PancakeSwapCAKE

Leituras Relacionadas

Behind the AI Scorecards Lies a Chinese 'Question Setter'

Behind the AI scorecards that dominate industry discussions—benchmarks like MMLU-Pro, MMMU, and MMMU-Pro—stands a Chinese-Canadian researcher: Wenhu Chen. As an assistant professor at the University of Waterloo and founder of the TIGER Lab, Chen has become a key "exam-setter" for evaluating large language and multimodal models. Chen first gained broader recognition with MMLU-Pro, a more challenging and stable update to the popular MMLU benchmark. As top models like OpenAI’s o3 began achieving near-perfect scores on the original MMLU, it became difficult to distinguish their true capabilities. MMLU-Pro introduced more complex reasoning questions, expanded answer choices, and filtered out ambiguous or simple items, effectively reintroducing differentiation among state-of-the-art models. His work on MMMU addressed the evaluation of multimodal models, requiring them to integrate visual information (like charts, diagrams, or tables) with textual knowledge across diverse academic subjects. Even the strongest models initially scored only around 56-59%, highlighting significant room for improvement in genuine multimodal reasoning. MMMU-Pro further refined this by preventing models from bypassing visual cues. Chen’s research focus has long been on complex information understanding and reasoning. His background—including a PhD at UC Santa Barbara, research at Google/DeepMind on Gemini, and now a role in Meta’s superintelligence lab—provides deep insight into model development and their potential weaknesses. His TIGER Lab also builds models (e.g., for video understanding and generation), ensuring his evaluation benchmarks are grounded in practical challenges. While AI headlines often spotlight company leaders and product launches, Chen’s work exemplifies the critical, behind-the-scenes contributions of researchers crafting the rigorous standards that define and drive progress in AI capabilities.

marsbitHá 8m

Behind the AI Scorecards Lies a Chinese 'Question Setter'

marsbitHá 8m

STRC Unpegged by 11%, Can Strategy's Perpetual Motion Machine Keep Turning?

STRC, the perpetual preferred stock of MicroStrategy, is experiencing a persistent de-pegging from its target par value of $100, with the discount recently widening to over 11%. This de-anchoring challenges the core design of STRC, which was intended as a stable, income-oriented security operating near $100. As a crucial funding engine for MicroStrategy's Bitcoin acquisition strategy, STRC's price reflects market confidence in the company's entire capital model. The company's "capital flywheel" relies on issuing STRC at or above $100 via an At-the-Market (ATM) program to raise cash for buying Bitcoin, thereby boosting company equity and theoretically supporting STRC's value. A monthly adjustable dividend mechanism was designed to maintain this peg. Despite raising the dividend to 11.5% and increasing payment frequency, the de-pegging persists. Market concerns extend beyond technical factors like leveraged arbitrage unwinding. Analysts point to MicroStrategy's limited cash reserves relative to its ~$1.7 billion annual dividend obligation for preferred shares. While the company counters that its vast Bitcoin holdings could cover decades of payments, this argument hinges on the potential need to sell Bitcoin—a shift from its longstanding "hodl" narrative. The company's recent sale of a small amount of BTC, framed as a test, amplified these liquidity and strategy concerns. If STRC remains discounted, impairing MicroStrategy's ability to raise cheap capital, fears may grow that the company could sell more Bitcoin to meet obligations. This scenario could transform MicroStrategy from a major market buyer into a potential seller, posing significant downside risk for Bitcoin. The re-pegging of STRC is thus a key indicator for the health of MicroStrategy's capital structure and its market impact.

Odaily星球日报Há 22m

STRC Unpegged by 11%, Can Strategy's Perpetual Motion Machine Keep Turning?

Odaily星球日报Há 22m

Silicon Valley's Most Sought-After New Role Has Emerged

Silicon Valley's New Most Wanted Job: The Rise of the Forward Deployment Engineer The AI industry is witnessing a significant shift. The focus has moved from developing cutting-edge models to deploying them effectively within enterprises. This has made the "Forward Deployment Engineer" (FDE) a critical and highly sought-after role at major firms like OpenAI, Anthropic, and Google. For the past three years, the industry prioritized model scientists. However, companies are now facing a harsh reality: purchasing powerful AI tools does not guarantee productivity gains or organizational change. The biggest hurdle is not the technology itself, but integrating it into complex legacy systems, workflows, and corporate cultures. This includes challenges like data silos, compliance requirements, and internal resistance. The FDE role, pioneered by Palantir Technologies, addresses this "last-mile" problem. FDEs are deployed on-site with clients for extended periods. Their job is to deeply understand the client's specific organizational structure, processes, and pain points, then tailor and implement the AI solution accordingly. They combine skills in technology, project management, and organizational change. A clear signal of this trend emerged in May 2026 when three AI giants made major moves. Anthropic launched a $1.5B joint venture for enterprise deployment. OpenAI formed an independent deployment subsidiary, DeployCo, with over $4B in commitments and acquired a deployment consultancy. Google Cloud's CEO publicly announced a large-scale recruitment drive for FDEs. This shift represents a fundamental change in the software business model: from selling tools to selling guaranteed outcomes. FDEs are the agents of this change, responsible for delivering a working system within the production environment, not just a demo. Real-world cases, such as challenges at Goldman Sachs (compliance barriers) and Target (internal cultural resistance), illustrate that the primary obstacles to AI adoption are organizational, not technical. An FDE's value lies in navigating these human and procedural complexities to facilitate a successful "AI migration." In essence, as core AI technology becomes more accessible and affordable, the true premium is shifting to the human expertise required to understand organizations and drive change—making the FDE role pivotal for the next phase of the AI revolution.

marsbitHá 22m

Silicon Valley's Most Sought-After New Role Has Emerged

marsbitHá 22m

When the World Cup Collides with Agents: From Web2 to Web3, How Are Wallets Evolving into Agentic Wallets?

World Cup as a Catalyst for Agentic Wallets: From Web2 to Web3 This article explores how the World Cup provides a real-world scenario for observing the evolution of digital wallets from simple asset managers towards "Agentic Wallets"—intelligent, AI-powered interfaces. Using the example of prediction markets like Polymarket, it illustrates how AI Agents can lower the barrier to Web3 interaction. Instead of navigating complex DApps, users can express intent in natural language (e.g., "I think Portugal will win") within platforms like Discord or web pages. The Agent then interprets this intent, finds the relevant market, and seamlessly guides the user through the on-chain transaction via their wallet. The core shift is from wallets as mere "function menus" for signing transactions to "intent interpreters" that understand user goals. The article highlights parallel developments in traditional finance, such as Mastercard's "Agent Pay" and WeChat Pay's AI tests, which focus on granting AI controlled, authorized, and auditable payment capabilities. This underscores a broader trend of AI entering the financial layer. However, the article emphasizes that the primary challenge for Agentic Wallets in Web3 is not automation but establishing clear security boundaries. Unlike traditional systems with chargebacks, on-chain transactions are often irreversible. Therefore, future wallets must ensure users retain ultimate control and comprehension. They need to transparently communicate an Agent's permissions, spending limits, authorized durations, and provide easy ways to pause or revoke access. The World Cup experiments represent early steps toward wallets that are not just applications but ubiquitous, intelligent interfaces that simplify Web3 while keeping users securely in control.

marsbitHá 1h

When the World Cup Collides with Agents: From Web2 to Web3, How Are Wallets Evolving into Agentic Wallets?

marsbitHá 1h

Options Don't Work in DeFi? Vitalik Might Not Agree

For years, the prevailing view has been that options struggle to gain traction in DeFi due to complexity, fragmented liquidity, and lack of natural demand compared to products like perpetual futures. However, a recent algorithmic stablecoin design proposed by Vitalik Buterin presents a different perspective, using options not as a standalone trading product, but as foundational infrastructure for other financial instruments. In this design, one unit of ETH is split into two components: a "stable" side (P) that retains value up to a specified strike price, and an "upside" side (N) that captures all appreciation above that strike. Combined, they always equal one ETH, eliminating debt, margin, and liquidation risks inherent in typical collateralized debt position (CDP) stablecoins. The stable component essentially mimics the payoff of a covered call option. To function as a stablecoin, this structure requires continuously rolling deep in-the-money calls, which introduces challenges like rollover slippage, predictable transaction flow vulnerable to front-running, and persistent liquidity needs. A core hurdle is finding consistent buyers for the leveraged ETH upside exposure (N). While it offers leverage without funding rates or liquidation, it must compete with simpler alternatives like direct call options or perpetuals. The system's scalability depends on a sustained demand for this specific form of leverage. The author draws parallels to their experience with Rysk, where earlier versions of DeFi options protocols struggled. The breakthrough came with Rysk V12, which aligns incentives: asset holders generate yield by selling covered calls against their holdings, while market makers efficiently acquire the desired option exposure. This demonstrates that options can find product-market fit when embedded as a risk distribution and pricing engine within structured products, stablecoins, or yield-generating assets, rather than marketed as a complex direct trading instrument. Vitalik's proposal reinforces this architectural approach—using fully collateralized, non-custodial, and physically settled options as a fundamental building block. The real opportunity for options in DeFi may lie not in becoming the next perpetual swap, but in powering the next generation of on-chain financial products.

marsbitHá 2h

Options Don't Work in DeFi? Vitalik Might Not Agree

marsbitHá 2h

Trading

Spot

Futuros

Artigos em Destaque

Como comprar BANANA

Bem-vindo à HTX.com!Tornámos a compra de Banana Gun (BANANA) simples e conveniente.Segue o nosso guia passo a passo para iniciar a tua jornada no mundo das criptos.Passo 1: cria a tua conta HTXUtiliza o teu e-mail ou número de telefone para te inscreveres numa conta gratuita na HTX.Desfruta de um processo de inscrição sem complicações e desbloqueia todas as funcionalidades.Obter a minha contaPasso 2: vai para Comprar Cripto e escolhe o teu método de pagamentoCartão de crédito/débito: usa o teu visa ou mastercard para comprar Banana Gun (BANANA) instantaneamente.Saldo: usa os fundos da tua conta HTX para transacionar sem problemas.Terceiros: adicionamos métodos de pagamento populares, como Google Pay e Apple Pay, para aumentar a conveniência.P2P: transaciona diretamente com outros utilizadores na HTX.Mercado de balcão (OTC): oferecemos serviços personalizados e taxas de câmbio competitivas para os traders.Passo 3: armazena teu Banana Gun (BANANA)Depois de comprar o teu Banana Gun (BANANA), armazena-o na tua conta HTX.Alternativamente, podes enviá-lo para outro lugar através de transferência blockchain ou usá-lo para transacionar outras criptomoedas.Passo 4: transaciona Banana Gun (BANANA)Transaciona facilmente Banana Gun (BANANA) no mercado à vista da HTX.Acede simplesmente à tua conta, seleciona o teu par de trading, executa as tuas transações e monitoriza em tempo real.Oferecemos uma experiência de fácil utilização tanto para principiantes como para traders experientes.

141 Visualizações TotaisPublicado em {updateTime}Atualizado em 2026.06.02

Discussões

Bem-vindo à Comunidade HTX. Aqui, pode manter-se informado sobre os mais recentes desenvolvimentos da plataforma e obter acesso a análises profissionais de mercado. As opiniões dos utilizadores sobre o preço de BANANA (BANANA) são apresentadas abaixo.