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 a 2026-04-27Actualizado a 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

Criptos en tendencia

CitreaCTR

wrapped stUSDTWSTUSDT

Velodrome FinanceVELODROME

BrevisBREV

ZRX（0X）ZRX

PancakeSwapCAKE

Lecturas Relacionadas

No Sales Team, $20 Million in Revenue: How Did AI Employee Viktor Win Over 30,000 Companies?

The AI employee Viktor, developed by a team with DeepMind background, has achieved $20 million in annual revenue without a traditional sales team, serving over 30,000 companies. Its core innovation lies in positioning itself as a "Tier 3 AI Coworker" capable of "end-to-end execution and delivery of results," moving beyond the "draft and wait for human completion" model of typical AI assistants. Users can simply mention Viktor in Slack or Microsoft Teams using natural language commands, and it autonomously performs tasks like pulling sales data from a CRM, generating reports, or even cross-tool operations like creating board meeting PPTs by aggregating data from six different sources. Key to its growth is a pure Product-Led Growth (PLG) model, eliminating complex implementation cycles and per-seat licensing. Instead, it charges based on task credits or consumption, lowering the trial barrier with a $100 free credit offer and no credit card required. This enabled viral, bottom-up adoption within organizations. Viktor's interaction paradigm removes the barrier of prompt engineering, allowing non-technical employees to delegate complex workflows seamlessly. It also features proactive, automated task execution (e.g., overnight bookkeeping, scheduled reports) based on triggers, effectively embedding AI as an automated "process layer" within business operations. However, its expansion into Microsoft Teams—a platform with 320 million users—highlights challenges. Large enterprises require stringent IT compliance, security reviews (e.g., SOC 2), and governance, potentially hindering the frictionless, user-driven adoption that succeeded in Slack. Additionally, the "black box" nature of its autonomous decision-making raises concerns about operational risks, data integrity, and the need for robust audit logs and permission controls. Balancing efficiency gains with security and trust remains a critical hurdle for Viktor and similar AI agents aiming to become core enterprise infrastructure.

marsbitHace 25 min(s)

No Sales Team, $20 Million in Revenue: How Did AI Employee Viktor Win Over 30,000 Companies?

marsbitHace 25 min(s)

Interview with CoreWeave Co-founders: AI Demand Seems to 'Intensify' Every Day

An Interview with CoreWeave Executives: AI Demand Seems to 'Intensify' Every Day In an interview, CoreWeave executives highlight a structural shift in AI infrastructure demand. While GPU availability remains crucial, the primary bottlenecks are evolving to include powered data center shells, skilled labor (like electricians), and complex supply chain execution. They note that AI demand, particularly for agentic AI and reasoning models, continues to intensify daily, accelerating since Q1 2024. This demand is driving a need for more balanced infrastructure. CoreWeave is redesigning data centers to allocate more space for storage and CPUs alongside GPUs, with significant interest in Nvidia's upcoming Vera CPUs. The company, serving top AI labs and hyperscalers, emphasizes its client-driven model, building precisely to customer specifications. CoreWeave attributes its competitive edge to proven execution, performance, and a mature platform for AI deployment. Pricing is structured to pass component cost increases (e.g., for HBM memory) to customers, protecting margins. Looking ahead, they anticipate Vera Rubin platform deployments to begin meaningfully in late 2025, with a major ramp throughout 2027, mirroring the Blackwell (GB) series rollout pattern. The competition is shifting from merely acquiring chips to holistic engineering and delivery capability.

marsbitHace 43 min(s)

Interview with CoreWeave Co-founders: AI Demand Seems to 'Intensify' Every Day

marsbitHace 43 min(s)

Manus Buyback Plan Emerges: Chinese Investors Plan to Repurchase Equity with $2 Billion, Path to Hong Kong IPO Becomes Clearer

According to a report by The Information, early Chinese investors of Manus, including Tencent, Sequoia Capital China, and ZhenFund, are planning to repurchase the company from Meta for $2 billion—the same price Meta paid in its acquisition last December. This move is a direct response to the Chinese government's prohibition of the foreign acquisition in April. As part of the repurchase plan, Manus is considering establishing a Sino-foreign joint venture within China. This structure is seen as a way to ensure regulatory compliance for its Chinese investors and to pave the way for a future IPO in Hong Kong. Notably, U.S. investor Benchmark will not participate in the buyback, which will concentrate ownership even more among Chinese capital. Since its acquisition by Meta, Manus's business has grown rapidly, with its annualized revenue run rate reportedly increasing four-to-fivefold to $400-$500 million in roughly six months. This strong growth underpins the investors' willingness to repurchase at the original price. Financially, the forced unwinding of the deal may benefit the early investors, allowing them to regain equity at a cost far below the company's current implied valuation, with the added prospect of an independent future listing. However, specific terms of the repurchase, including funding proportions and the joint venture's equity structure, are still under negotiation. This "repurchase-joint venture-Hong Kong IPO" approach could serve as a reference model for other Chinese AI startups navigating cross-border M&A regulations.

marsbitHace 52 min(s)

Manus Buyback Plan Emerges: Chinese Investors Plan to Repurchase Equity with $2 Billion, Path to Hong Kong IPO Becomes Clearer

marsbitHace 52 min(s)

STRC Loses Peg by 11%, Can Strategy's Perpetual Motion Machine Keep Running?

The article discusses the significant and concerning depegging of MicroStrategy's (MSTR) preferred stock, STRC. Designed to trade near its $100 target par value, STRC has recently fallen sharply, reaching a low of $83.26 and closing at $88.59, representing an over 11% discount. STRC is a core component of MicroStrategy's financial strategy. As a perpetual preferred stock, it allows the company to raise capital through an "at-the-market" (ATM) issuance program without diluting common shareholders (MSTR). This capital is primarily used to purchase Bitcoin, creating a "capital flywheel": issuing STRC → raising cash → buying BTC → increasing net assets → supporting STRC's value. The flywheel's operation depends on STRC maintaining its $100 price. To enforce this, MicroStrategy employs a dynamic dividend mechanism, recently raising the rate to 11.5% and increasing payout frequency. However, this has failed to halt the depegging, indicating market concerns extend beyond yield. Analysts cite two main reasons. First, technical factors like forced liquidations from leveraged arbitrage trades may have exacerbated the sell-off. Second, and more fundamentally, is waning confidence in MicroStrategy's financial resilience. A JPMorgan report highlighted the company's limited cash relative to its ~$1.7 billion annual dividend obligation, raising liquidity concerns. While MicroStrategy counters that its massive Bitcoin holdings provide decades of coverage, this argument relies on the potential need to sell BTC—a departure from its long-standing "never sell" narrative. The company's recent sale of a small amount of Bitcoin for "testing," despite being framed as minor, has intensified these fears. The persistent depegging threatens to cripple MicroStrategy's primary funding channel. If STRC remains discounted, the company's ability to fund further Bitcoin purchases weakens. Should cash reserves dwindle while financing is constrained, the market may increasingly price in the risk of MicroStrategy becoming a forced seller of Bitcoin to meet obligations. This shift from a major marginal buyer to a potential seller could pose significant downside risk to the broader Bitcoin market.

链捕手Hace 1 hora(s)

STRC Loses Peg by 11%, Can Strategy's Perpetual Motion Machine Keep Running?

链捕手Hace 1 hora(s)

Altcoins Face Deepest Spot Sell Pressure Since 2020, CryptoQuant Data Shows

Altcoins are enduring one of their most severe periods of spot-market selling pressure since 2020, according to CryptoQuant data, with a cumulative buy/sell volume gap of roughly $209 billion. This reflects weak retail demand, a rotation into safer assets like Bitcoin, Ethereum, and stablecoin yield products, and broad investor caution. While such extreme selling can set the stage for a contrarian opportunity by showing one-sided positioning, the data does not yet signal an immediate reversal or altcoin season. The market remains in a defensive state, lacking the sustained spot accumulation and broad sector improvement needed for a confirmed bullish turn.

bitcoinistHace 1 hora(s)

Altcoins Face Deepest Spot Sell Pressure Since 2020, CryptoQuant Data Shows

bitcoinistHace 1 hora(s)

Trading

Spot

Futuros

Artículos destacados

Cómo comprar BANANA

¡Bienvenido a HTX.com! Hemos hecho que comprar Banana Gun (BANANA) sea simple y conveniente. Sigue nuestra guía paso a paso para iniciar tu viaje de criptos.Paso 1: crea tu cuenta HTXUtiliza tu correo electrónico o número de teléfono para registrarte y obtener una cuenta gratuita en HTX. Experimenta un proceso de registro sin complicaciones y desbloquea todas las funciones.Obtener mi cuentaPaso 2: ve a Comprar cripto y elige tu método de pagoTarjeta de crédito/débito: usa tu Visa o Mastercard para comprar Banana Gun (BANANA) al instante.Saldo: utiliza fondos del saldo de tu cuenta HTX para tradear sin problemas.Terceros: hemos agregado métodos de pago populares como Google Pay y Apple Pay para mejorar la comodidad.P2P: tradear directamente con otros usuarios en HTX.Over-the-Counter (OTC): ofrecemos servicios personalizados y tipos de cambio competitivos para los traders.Paso 3: guarda tu Banana Gun (BANANA)Después de comprar tu Banana Gun (BANANA), guárdalo en tu cuenta HTX. Alternativamente, puedes enviarlo a otro lugar mediante transferencia blockchain o utilizarlo para tradear otras criptomonedas.Paso 4: tradear Banana Gun (BANANA)Tradear fácilmente con Banana Gun (BANANA) en HTX's mercado spot. Simplemente accede a tu cuenta, selecciona tu par de trading, ejecuta tus trades y monitorea en tiempo real. Ofrecemos una experiencia fácil de usar tanto para principiantes como para traders experimentados.

142 Vistas totalesPublicado en 2024.12.11Actualizado en 2026.06.02

Discusiones

Bienvenido a la comunidad de HTX. Aquí puedes mantenerte informado sobre los últimos desarrollos de la plataforma y acceder a análisis profesionales del mercado. A continuación se presentan las opiniones de los usuarios sobre el precio de BANANA (BANANA).