The Builders Stage is making a comeback at TechCrunch Disrupt 2026, uniting founders, startup professionals, and investors for insightful discussions on the essentials of constructing and expanding successful enterprises.

Listen to startup and venture leaders influencing the tech landscape, such as Grant Lee, CEO and co-founder of Gamma; Leah Solivan, founder and general partner at Precedent.vc; Robby Stein, VP of Product at Google; and more. Through open dialogues and practical case studies, speakers will provide actionable insights on fundraising, talent acquisition, market strategies, AI, and the key operational choices that drive startup development.

Join over 10,000 founders, investors, startup professionals, and tech leaders at Moscone Center in San Francisco from October 13-15. Register now to save up to $330 before ticket prices rise.

Designed for founders ready to expand

Starting a startup is one challenge; creating a scalable company is another. The Builders Stage is one of six industry-centric stages at Disrupt 2026, focused on aiding founders in tackling growth challenges, from capital raising and recruiting top talent to constructing market strategies and preparing for the transition from seed to Series A.

Each session offers practical tactics you can immediately implement, alongside chances to interact directly with speakers during live Q&A. Get your ticket to Disrupt 2026 today and save up to $330 before prices increase.

Without further delay, here’s your initial preview of the Builders Stage schedule, with more speakers and sessions to be revealed as the event draws nearer.

Builders Stage schedule

Winning Without Building AI

Featuring Shan Shan, Investment Manager, Baillie Gifford; and Yuri Sagalov, Managing Director at General Catalyst

While AI may dominate the venture landscape, many lasting businesses won’t necessarily be those that offer AI models or services. This discussion caters to founders vying for attention in an AI-centric market. Panelists will unpack the crucial aspects: efficient growth, retention, revenue integrity, and methodical execution, explaining why core principles, rather than hype, still foster breakthrough enterprises.

When OpenAI Sends You Your Roadmap

Featuring Michel Tricot, CEO and Co-founder, Airbyte; Rob Toews, Partner, Radical Ventures; and Linda Tong, CEO, Webflow

Many AI founders worry: What happens if OpenAI or Anthropic unveils a competing product? Even strong offerings risk being classified as features of larger companies. This session investigates areas of defensibility and what steps founders can take if confronted with competition from fast-evolving AI giants.

Pre-Seed Success Without a Product

Featuring Puneet Agarwal, Managing Partner, True Ventures; Austin Clements, Managing Partner, Slauson and Co; and Sandhya Venkatachalam, Founder and Managing Partner, Axiom Partners

Founders are increasingly expected to secure funding before launching a product. At the pre-seed phase, investors often invest based on narrative, conviction, and founder-market fit. This session outlines how to establish credibility prior to revenue generation so that investors are willing to issue that first investment.

Transitioning from MVP to Millions of Users: How Product Choices Change at Scale

Featuring Robby Stein, VP, Product, Google

The instincts necessary for creating your initial minimum viable product can lead to challenges at a billion-user scale. In this conversation, Robby Stein discusses how decision-making around product evolves as every update affects billions of users. Learn how teams balance speed with trust and innovation with dependability at one of the largest product organizations globally.

Recruiting When AI Acts as a Co-Founder

Featuring Josh Reeves, CEO and Co-founder, Gusto; additional speakers forthcoming

Early-stage firms are no longer merely incorporating AI; they are also hiring it. With AI agents taking over engineering, support, and operational roles, the definition of a founding team is undergoing transformation. This session delves into how founders determine which responsibilities should rest with humans versus what can be automated through AI, and how high-growth startups are developing hybrid teams without compromising speed, responsibility, or organizational culture.

M&A as an Early-Stage Strategy

Featuring Karl Alomar, Managing Partner, M13; Aklil Ibssa, Head of Corporate Development and M&A, Coinbase; and Lindsey Mignano, Founder, Mignano Law Group

Savvy founders today aren’t solely building for IPOs; they are also considering potential acquisitions from the outset. As exit strategies change and available capital narrows, understanding M&A early has become an essential competitive edge. This session outlines how founders can set the stage for such possibilities through product strategy and collaborations, detailing how significant startup achievements happen, even for smaller firms.

Series A in 2027

Featuring Jahanvi Sardana, Partner, Index Ventures; Shailendra Singh, Managing Director, Peak XV; and Janelle Teng Wade, Partner, Bessemer Venture Partners

Securing Series A funding is becoming increasingly challenging, as VCs become more discerning. For founders aiming to raise funds in the next one to two years, this session outlines what “fundable” will look like by 2027. Discover how leading investors are redefining the important metrics, teams, and traction, what outdated fundraising strategies are no longer effective, and how companies can differentiate themselves in the upcoming funding cycle.

The 90-Day GTM: Why $0–$10M ARR is the New Standard (and How to Achieve It)

Featuring Ryan Meadows, Chief Revenue Officer, Lovable; and Tomasz Tunguz, General Partner and Founder, Theory Ventures; more speakers to be confirmed

The definition of traction has evolved. What once took years is now anticipated within months, and achieving $0–$10M ARR is becoming the new primary benchmark for early-stage companies. This discussion examines how AI-driven execution, faster distribution channels, and evolving expectations from investors are compressing go-to-market timelines, alongside the tactical approaches founders need to accelerate revenue in the crucial first 90 days and stand out quickly.

The True Tokenmaxxing: How Leading AI Companies Operate in a Multi-Model World

With Mo Jamma, Partner, Capital G; Zuzanna Stamirowska, CEO and Co-founder, Pathway; and additional speakers to be announced

The pace of advancement is faster than any single model can match, and the teams that are developing the most effective AI solutions are increasingly utilizing multiple models instead of relying solely on one. This panel gathers founders and operators central to this transition to discuss their methods for evaluating new models, managing scale-related costs and reliability, and designing products that can adapt as rapidly as the underlying technology changes.

PMF Warning Signs: Identifying Genuine Product-Market Fit

Featuring Rajeev Dham, Partner, Sapphire Ventures; Rahul Vohra, Founder & Head of Superhuman Mail; and upcoming speakers

During an AI hype phase, indicators of product-market fit can be misleading and difficult to evaluate. Founders often confuse early enthusiasm, spikes in usage, and pilot successes for sustainable traction. This session clarifies what illusory PMF looks like, how investors and operators differentiate authentic retention from adoption driven by hype, and the signals that reveal whether a company genuinely attracts customers or merely experiences temporary momentum.

The Zero-to-1K Strategy: Securing Your Initial 1,000 Customers Without a Marketing Budget

Featuring Grant Lee, CEO and Co-founder, Gamma and Leah Solivan, Founder and General Partner, Precedent.vc

Gaining early customers is not about marketing expenditures; it’s rooted in founder-driven distribution and relentless execution. Startups in the initial phase often lack funds, branding, or scale, relying instead on urgency and ingenuity. This session outlines strategies founders utilize to attract their first customers through community engagement, product-led growth, founder-led sales, strategic outbound efforts, and organic word-of-mouth.

Yes, Being a Founder is Difficult: A Candid Discussion

Featuring Nell Daly, Co-Founder and Managing Partner, Revenge Capital; David H. Rosmarin, Associate Professor, Harvard Medical School; and Jack Withinshaw, Co-founder and Chief Commercial Officer for Airspeeder

Building a company is as psychologically demanding as it is strategically challenging, and many narratives surrounding founders overlook this reality. In this honest dialogue, founders and mental performance experts reveal the hidden toll of high-growth settings, from burnout and decision fatigue to the personal strain of continuous pressure, while sharing the systems, practices, and mental frameworks that help leaders thrive and maintain high performance.

You Have a Successful Product. How Will Your Company Repeat That Success?

Featuring Filip Kaliszan, CEO and Co-Founder, Verkada; and additional speakers to be announced

Many startups reach an impasse because they produce one excellent product rather than developing a replicable multi-product framework. Join a venture capitalist and two founders as they outline a thorough operational playbook for wise capital distribution, establishing systematic innovation, and crafting a compounding “Second Act” before the primary product’s growth trajectory stabilizes.

Recruitment, Compensation, and Culture in Today’s Highly Competitive Market

Featuring Matt Birnbaum, Founder, Wylder.co; Atli Thorkelsson, VP, Talent Network, Redpoint Ventures; and more speakers to be confirmed

Unquestionably, the rise of AI startups has complicated hiring and retention across the tech industry. From contending for AI expertise to secondary sales, founders are reevaluating the human framework of their businesses. As hiring practices, incentives, and employee expectations evolve rapidly, this session discusses how companies are modifying compensation, organizational culture, and team-building methods to attract and retain top talent in an ever-altering startup climate.

How to Foster Viral Growth and Leverage It

Featuring Zach Yadegari, Founder, Cal AI

Startups can achieve viral success overnight, yet sustaining this momentum presents an entirely different obstacle. During this conversation, Zach Yadegari discusses how Cal AI managed rapid growth, product pressures, and the realities of functioning in a distribution-focused market. Discover the insights behind converting sudden attention into sustainable retention and enduring company development.

The High-Conviction Assessment: Lessons from the Battlefield

Featuring Alexa Von Tobel, Inspired Capital and more speakers to be confirmed

What differentiated breakout companies from others at Disrupt 2026? In this transparent recap, Battlefield judges analyze trends and founder characteristics that stood out in real-time, from evolving investor expectations to the narratives that resonated most this year. Additionally, the discussion will address the evolution of startup storytelling and the realities of sustaining momentum and navigating the pivotal 12 months following a significant launch, funding round, or Battlefield participation.

Engage in the discussions and forge connections at Disrupt

If you are prepared to innovate smarter, accelerate growth, and learn from the key figures shaping the future of startups, secure your ticket to TechCrunch Disrupt 2026 now and save up to $330 before the rates rise.

Meta has invested billions in AI development and the expansion of data centers to facilitate it. However, it appears the company is now gearing up to utilize these data centers for more immediate profit.

On Wednesday, Bloomberg disclosed that Meta is crafting plans for a cloud infrastructure venture, offering access to both AI computing capabilities and models. This initiative would position it in competition with major cloud service providers like Amazon Web Services, Google Cloud, and Microsoft Azure. 

Meta’s choice to monetize surplus computing power comes just weeks after SpaceX, through xAI, revealed comparable plans. In early May, SpaceX entered into an agreement with Anthropic to purchase all compute resources at its Colossus 1 data center. SpaceX has since established similar agreements with Google and Reflection AI. The fact that Meta is undertaking a similar approach suggests that the leaders in the AI race may not solely be those offering superior models and services but rather those possessing the data centers.

This assumes that the demand for computing remains steady and that data centers maintain their value. Some critics have raised concerns that the rush to expand AI infrastructure could result in a bubble heavily reliant on rapidly depreciating chips. Others have questioned whether AI firms can produce sufficient end-user revenue to warrant the trillion-dollar investments. 

Despite these concerns, Meta has continued to heavily invest in AI computing infrastructure. As of the end of the first quarter, Meta has pledged $182.9 billion for AI infrastructure enhancements in the upcoming years, which includes large-scale ongoing projects in Louisiana and Ohio. The Ohio initiative, which Zuckerberg noted would be equivalent in size to Manhattan, is anticipated to launch this year.

In contrast to Google and OpenAI, Meta has not witnessed substantial interest in its proprietary AI models and services. Meta does not specify its revenue derived from Meta AI or from its Llama open-weight AI model family in its earnings reports, and executives have largely highlighted AI’s internal corporate applications in public comments. This may indicate that Meta’s AI projects have yet to provide a significant standalone revenue stream. 

To recoup some of its considerable investment, Meta might emulate CoreWeave’s business approach by selling access to “raw” compute power, according to Bloomberg. The outlet further reported that Meta is mulling over following AWS’s example and providing access to various AI models—including its newly introduced closed-weight model, Muse Spark—hosted on its AI infrastructure.

This new business initiative will be part of a new effort reportedly named Meta Compute, directed by infrastructure head Santosh Janardhan, Meta Superintelligence Labs leader Daniel Gross, and president Dina Powell McCormick.

The report affirms Zuckerberg’s statements from May, indicating that creating a Meta cloud computing business is “definitely on the table” as a strategy to recoup some of the significant investments in its quest to develop AI “superintelligence.”

TechCrunch has contacted Meta for a response.

Vinton Cerf will resign from his position as Google’s chief internet evangelist next week, signaling the end of one of the most significant careers in the history of technology.

During a video address at the Open Frontier conference organized by the Laude Institute, Cerf was honored by Dave Patterson, a professor at UC Berkeley renowned for co-developing RISC processor architecture.

“Vint… has been with Google for over 20 years, and he is retiring a week from now, so I believe we should give him a round of applause for a commendable career,” Patterson remarked, drawing cheers from the audience.

Google did not provide a comment by the time of publication.

Cerf, 83, along with collaborator Robert Kahn, is acknowledged as one of the creators of the networking protocols that underlie the internet as we recognize it today. His contributions to the development and dissemination of TCP/IP — the fundamental framework that enables communication among various computer networks — starting in the 1970s have earned him numerous honorary degrees, the Presidential Medal of Freedom, and a Turing Award, among other accolades.

Since 2005, Cerf has held the title of vice president and chief internet evangelist at Google. (At this juncture, we can confidently assert the internet has been fully evangelized, for better or worse.)

Cerf participated in a panel discussion alongside other computer scientists recognized for their contributions to sustainable open-source projects, including Patterson; François Chollet, the creator of the Keras deep-learning library and co-founder of Ndea; John Ousterhout, the Stanford computer scientist who developed the Tcl programming language and co-founded Electric Cloud; and Matei Zaharia, co-founder and chief technologist of Databricks. They shared insights on building open-source systems that thrive — a topic that is increasingly important as founders invest in open infrastructure for the next generation of AI products.

A significant portion of the conference discussions centered on the issues surrounding the centralization of advanced models within a limited number of well-funded labs, contrasting with the decentralized nature of the open internet that enabled Cerf’s protocols to endure. Nevertheless, Cerf anticipated that the emergence of AI agents — software capable of autonomous action and collaboration with other software — would compel tech companies to revert to standardized protocols.

“The agentic model of AI, characterized by multiple agents from diverse sources interacting, will necessitate composability along with a demand for interoperability and standardization,” Cerf stated.

If he is correct, the firms that establish these interoperability standards early on could gain significant influence over the workings of the agentic economy — an interplay reminiscent of the early internet protocol conflicts.

While other panelists suggested that natural language communication between LLM agents would suffice, Cerf foresaw the need for formal standards.

“I don’t believe English will be the optimal choice. Its flexibility brings ambiguity, and I think precision in inter-agent communication will be crucial. An agent must ensure that the other agent comprehends what it is they have just agreed to do together,” Cerf noted.

“Recall the classic telephone game where you wish you’d whispered directly in someone’s ear and by the time it got to the tenth person, the message had completely changed? Picture a bunch of agents communicating in natural language — that’s quite alarming.”

In a lighter moment, Patterson reminisced about meeting Cerf, who is noted for his array of three-piece suits, during his graduate studies in the 1970s.

“He’s always been the best-dressed computer scientist I’ve encountered,” Patterson remarked. “When I recall Vint, he was a grad student sporting a shirt and tie in the ’70s.”

“That is absolutely true,” Cerf responded. “I even had a vest, and for some reason, I always wanted to stand out, opting for unique attire instead of long hair or piercings — I thought dressing differently was the way to do it.”

The U.S. has removed the requirement for Anthropic to secure a license prior to exporting its Mythos and Fable models internationally, a stipulation that had effectively restricted public access to what are regarded as the most sophisticated AI models available to date.

The AI research lab announced that it would start reinstating access to the models on Wednesday, July 1.

On June 12, the U.S. administration classified the products as export-restricted technologies, meaning they could no longer be accessible to foreign individuals without specific authorization. Adhering to this regulation became unfeasible at scale, compelling Anthropic to cease public accessibility to the models entirely.

Following weeks of discussions, Secretary of Commerce Howard Lutnick stated that Anthropic “has consented to actively identify and mitigate security risks linked to the models; to collaborate closely with the U.S. government on protocols, standards, and releases for Mythos, Fable, and forthcoming models; and to notify the U.S. government of any malicious activities.”

Anthropic had previously made a public commitment to undertake much of this voluntarily, several months before the establishment of the export regulation. This is partly why cybersecurity professionals were doubtful about the initial restrictions. To them, the ban appeared more as a means of leverage, a method for the Trump administration to penalize Anthropic for its leaders’ open criticism of how the government, including the president’s political rivals, might leverage the technology.

Mythos was initially available only to a limited number of organizations starting in April to mitigate concerns regarding its capacity to detect and manipulate vulnerabilities in software, while a version named Fable was introduced to the public in June with enhanced security measures.

Nonetheless, as Asian AI firms began unveiling their own models nearing Mythos-level capabilities — including Fugu and Tulongfeng — the U.S. government faced pressure to relax its restrictions on Anthropic to maintain the competitiveness of American AI in the global market.

Last week, Lutnick authorized Mythos for release to selected customers approved by the White House. OpenAI’s recent models were also made available to a group of organizations sanctioned by the Trump administration, rather than the general public.

The Trump administration’s unpredictable stance on AI policy has resulted in a lack of clarity for companies throughout the industry regarding the governing rules for future model releases. An executive order enacted in June, which indicated an intent to evaluate models prior to release, received criticism from prominent analysts such as Dean W. Ball, who recently took on a policy role at OpenAI.

Wayve, a self-driving technology startup based in the U.K., is allowing its employees to divest a portion of their vested equity. The $85 million tender offer — effectively a structured chance for employees to resell shares to investors — is being facilitated by both current and new investors at the company’s recent valuation of $8.5 billion.

This valuation was established in February when the nine-year-old firm secured a $1.2 billion Series D funding round led by Eclipse, Balderton, and SoftBank Vision Fund 2, with involvement from Ontario Teachers’ Pension Plan, Baillie Gifford, Microsoft, Nvidia, and Uber.

This marks Wayve’s second event for employee liquidity. The company had previously organized a tender offer during its $1.05 billion Series C funding round in May 2024.

Wayve’s initiative is indicative of a rising trend among AI startups. Instead of waiting many years for an exit strategy, companies are utilizing tender offers as a means of retention, providing employees with an incentive to remain rather than moving to competitors — or launching their own ventures — once their options mature.

Other startups that have recently executed employee tender offers include Decagon, which creates AI agents managing customer support for firms like Duolingo and Hertz; ElevenLabs, the AI voice-generation firm behind much of the internet’s synthetic voice and dubbing solutions; Linear, a favored project-management tool designed for software teams; and Clay, a sales and marketing automation platform that aids companies in researching and contacting prospects. (Clay has executed two tenders in the last nine months alone.)

These companies can offer employee liquidity mainly because investors are keen to acquire more equity in these rapidly growing ventures, even at a premium, betting that their value will increase in the future.

Wayve employs a self-learning methodology for its autonomous driving. Rather than depending on pre-existing, high-definition maps used by most self-driving technologies, its software represents an end-to-end neural network that learns to drive solely from data — akin to how humans learn to drive through practical experience, according to its founders.

Aiming for a “general-purpose” AI driver — one that could theoretically operate across different countries, vehicles, and road conditions — the company has more than doubled its workforce to 1,200 employees over the past year.

Wayve is planning robotaxi pilot launches in collaboration with Uber later this year, while also working to incorporate its AI software into Nissan’s next-gen driver-assist systems starting in 2027.

Clicks Technology, a startup gearing up to unveil its distinctive version of the BlackBerry smartphone, has showcased what it has planned in a new video released today.

The device, named the Clicks Communicator, was first unveiled at January’s Consumer Electronics Show (CES) in Las Vegas, targeting users who do substantial work on their phones, such as texting and emailing. It’s particularly designed to attract those who long for the BlackBerry’s physical keyboard, which many claim is superior for these tasks.

With a price tag of $499, the Communicator resembles what a contemporary BlackBerry might look like, featuring a display for viewing and responding to messages, along with a tactile, touch-sensitive keyboard underneath.

[embedded content]

However, the phone is more than just a BlackBerry imitation, as it brings forth innovation in various aspects. For example, its unique “Signal Light” feature includes a light-up button on the side of the device that can be customized with different colors and light patterns, indicating when messages are received from specific individuals, groups, or applications. 

It also features interchangeable back covers that can be easily removed and swapped, a 3.5 mm headphone jack, a physical SIM card slot (alongside an eSIM), expandable microSD storage (up to 2TB), and a tactile switch to toggle airplane mode on and off.

The Clicks Communicator may also attract a growing demographic of users seeking to disconnect from modern smartphones laden with addictive social apps and games. (The company has collaborated with the Niagara Launcher to ensure access to the Android apps it supports). With the Signal Light feature, you can safely disregard your phone unless a crucial notification pops up.

In the recent video, the company presents its pre-production hardware and internal software as a sneak peek of what’s to come when the phone starts shipping in Q4 of this year.

At CES, TechCrunch had the opportunity to experience the Communicator by handling a prototype that matched the size and weight of the device set to ship by the year’s end. It felt comfortable to hold, not overly light or heavy, and was easy to grip. The keys had a satisfying click, reminiscent of a BlackBerry, although the team intended to make some slight adjustments to the key pressure for improved performance for fast typers.

Future videos will delve deeper into specific features of the Clicks Communicator, including the Signal Light, Prompt Key, Message Hub, touch-sensitive keyboard, and additional functionalities.

Etched, a rival to Nvidia’s AI chip, released a progress update on Tuesday after TSMC successfully produced its chip earlier this year. The startup claims to have secured $1 billion in contract orders for its offering: complete systems powered by these chips.

Etched is presently engaged in customer testing of its initial product. These systems are referred to as “frontier inference clusters,” packages that consist of the chips along with specially designed racks and software, all intended to enable frontier models to perform inference more quickly, affordably, and efficiently than competitors, according to Etched. (Inference occurs after a user inputs a prompt — it is currently the largest bottleneck and cost driver for AI firms attempting to serve clients at scale, which is precisely why investors are focusing on those claiming to remedy this issue.)

Founded in 2022, Etched disclosed that it has raised a cumulative total of $800 million so far. The latest tranche was a previously undisclosed $500 million round completed in December, achieving a $5 billion post-money valuation, the firm noted.

The startup has garnered significant interest from a prominent group of investors, such as VentureTech Alliance, Jane Street, Hudson River Trading, Two Sigma, and Ribbit Capital. It has also attracted angel investments from AI luminaries including Andrej Karpathy, Geoffrey Hinton, Fei-Fei Li, Arthur Mensch, and Scott Wu. The cap table features billionaires Stanley Druckenmiller and Peter Thiel as well.

Although the startup’s press statement positions Tuesday’s announcement as Etched “emerging from stealth,” the co-founders — CEO Gavin Uberti and President Robert Wachen — have been discussing their chip ambitions with TechCrunch since 2024. Both left Harvard and became Thiel fellows to establish Etched, as Uberti informed TechCrunch at that time.

By 2024, Etched was already attracting investor interest, having raised over $125 million. However, during Patrick O’Shaughnessy’s “Invest Like the Best” podcast, the founders recounted their struggles in 2023 to engage investors — even with a comprehensive 30-page memo advocating that AI would ultimately require specialized chips instead of just general-purpose GPUs. Every major investor they approached declined. The company was reportedly operating on a month-to-month basis, nearing a cash shortfall in those initial stages.

Today’s investment landscape seems like an entirely different realm. Investors are pursuing every AI-related opportunity, particularly chip technology that accelerates inference. Competitor Cerebras achieved the first major IPO of the year, while AI chip manufacturer Groq recently raised $650 million. Hyperscalers like Amazon, Google, and Microsoft are all developing their own proprietary AI chips. Even OpenAI just announced its inaugural custom chip, created by Broadcom.

As the capacity for agentic functionality becomes an essential requirement among foundation model companies, Anthropic is unveiling Claude Sonnet 5, a more robust and agentic iteration of the lab’s mid-tier model. 

“It is capable of planning, utilizing tools such as browsers and terminals, and operating independently at a level that, only a few months back, necessitated larger and more costly models,” Anthropic noted in a blog entry. 

This perspective aligns with what OpenAI and Google have articulated regarding their latest releases. OpenAI’s GPT-5.6 Sol was introduced in preview last week as the firm’s most agentic model to date, enabling users to delegate tasks across subagents for extended autonomous projects. Google’s Gemini 3.5 Flash, launched in May, was promoted as a transition from a conversational chatbot to an agentic tool that strategizes, constructs, and iterates on actual tasks with minimal human involvement.

The emphasis on Sonnet 5 indicates that agentic capability is now the baseline expectation across all pricing levels. The distinguishing factor will no longer be who excels at agentic work, but rather how affordably and reliably they can execute it without human supervision.  

Sonnet 5 claims to deliver performance nearly on par with Opus 4.8, but at significantly reduced costs. Starting Tuesday, Claude Sonnet 5 will be the standard model for both free and Pro plans and is accessible for every subscription type.

At its launch, Sonnet 5 is set at $2 per million input tokens and $10 per million output tokens until August 31, after which the price will rise to $3 per million input tokens and remain at $10 per million output tokens. This pricing positions Sonnet 5 as less expensive than Opus 4.8, as well as OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Pro. (It remains pricier than Gemini 3.5 Flash.)

The updated model also shows considerable enhancements over its forerunner Sonnet 4.6, released in February, particularly in agentic performance metrics such as reasoning, tool usage, software development, and knowledge tasks, according to Anthropic. 

For instance, on one benchmark, Sonnet 5 achieves 63.2% in agentic coding, compared to Opus 4.8’s 69.2% and Sonnet 4.6’s 58.1%. On a knowledge work benchmark, Sonnet 5 actually slightly surpasses Opus 4.8, which is renowned for excelling at resolving complex issues like nuanced judgments and in-depth research. 

“Opus 4.8 remains the preferred model for achieving higher accuracy on these challenges, but Sonnet 5 offers developers more affordable options that are significantly superior to those previously available,” Anthropic states. “Between Sonnet 5 and Opus 4.8, users can modify the effort level to strike the right balance between cost and efficacy.”

Testers referenced in the blog post indicate that Sonnet 5 also excels at completing intricate tasks where earlier model versions would have faltered and “reviews its own output without being explicitly instructed.”

“We assigned Claude Sonnet 5 a two-part task — updating Salesforce account tiers and sending a launch announcement to enterprise contacts — and it completed the entire process,” Daniel Shepard, a senior engineer at Zapier, commented. “That would have stalled midway before. For everyday automation, it’s an obvious choice.”

In terms of safety, Sonnet 5 also exhibits a lower frequency of “undesirable behaviors” such as collusion with misuse and deception compared to its predecessor, making it safer for use in agentic environments. It is more adept at declining harmful requests and evading hijacking attempts in prompt-injection assaults. Additionally, it hallucinates and engages in sycophantic behavior less frequently than Sonnet 4.6.

However, it does not match the capabilities of Opus 4.8 and Claude Mythos Preview regarding misaligned behavior. “Evaluations indicate that it significantly lags in the ability to execute dangerous cybersecurity tasks compared to our current Opus models,” the blog post states.

Lovable co-founder Fabian Hedin remarked that Claude Sonnet 5 “consistently and effectively declines unsafe requests.”

“At Lovable, we’re equipping millions of creators with powerful tools,” Hedin mentioned. “A model that knows when to decline is just as critical as one that understands how to construct.”