frontier AI models are not too big because the technology is complex and too big because the training data is garbage

Updates between versions are bolded.

See updates to the core system prompts on claude.ai and the Claude iOS and Android apps.

This prompt is periodically updated to improve Claude's responses.

These system prompt updates do not apply to the Claude API.

The system prompt also encourages certain behaviors, such as always providing code snippets in Markdown.

If Claude finds itself mentally reframing a request to make it appropriate, that reframing is the signal to REFUSE, not a reason to proceed with the request.

But that comes with a new risk: While scripted conversations can't really go off the rails, ones generated by AI certainly can. Some popular AI toys have, for example, talked to kids about how to find matches and knives.

Amazon, Toyota, and GXO (a logistics giant with customers like Apple and Nike) have all deployed it—making it one of the first examples of a humanoid robot that companies see as providing actual cost savings rather than novelty.

In 2025, Google DeepMind further fused the worlds of large language models and robotics, releasing a Gemini Robotics model with improved ability to understand commands in natural language.

The solution is called domain randomization. You essentially create millions of simulated worlds that all vary slightly and randomly from one another.

Companies and investors put $6.1 billion into humanoid robots in 2025 alone, four times what was invested in 2024.

Developers will be forced to look elsewhere, from smaller models to on-premise deployments, until energy infrastructure & data center buildouts catch up, which could take years.

This imbalance will inevitably drive prices higher as demand compounds against a fixed supply.

The age of abundant AI is over, & it will remain so for years.

Anthropic has limited its newest model to roughly forty organizations.

This feels like a preview of where SaaS economics are heading. The companies that built big orgs on the assumption of steady seat expansion are going to find themselves competing with products built by tiny teams inside the frontier labs.

Figma has close to 2,000 employees - not all working on product engineering of course. I really doubt Anthropic even needed 10 to build Claude Design.

It's also worth noting that a lot of the things that would traditionally lock a company like Figma in stop working as well in an agent-first world.

Figma is effectively funding a competitor - and the more AI usage Figma has - the more money they send over to Anthropic for the tokens they use.

A system that can look up any fact has not been forced to find structure. It has not been forced to generalize. The lossy compression that makes training so powerful, the mechanism that turns raw data into transferable representations, is exactly what we shut off the moment we deploy.

The filing cabinet keeps getting bigger. But a bigger filing cabinet is still a filing cabinet. The breakthrough is letting the model do after deployment what made it powerful during training: compress, abstract, and learn.

The current separation between training and deployment is not just an engineering convenience – it is a safety, auditability, and governance boundary. Open it, and several things break at once.

The irony is that the very mechanism that makes LLMs powerful during training (e.g. compressing raw data into compact, transferable representations) is exactly what we refuse to let them do after deployment.

Large language models live in a similar perpetual present. They emerge from training with vast knowledge frozen into their parameters but they cannot form new memories – cannot update their parameters in response to new experience.

Such tools promise to turn coders into project managers, letting them delegate and oversee many more tasks than they could cope with by themselves.

But the real power of agents comes when they can work as a team. Instead of lone-wolf bots carrying out single tasks, such as using a browser to make a restaurant reservation or sending you a summary of your inbox, new tools can yoke together multiple agents, give each of them a different job, and orchestrate their behaviors so that they all pull together to complete more complex tasks than an individual agent could do by itself.

Think of multi-agent systems as the new assembly lines. Henry Ford's innovation upended entire industries last century. In theory, networks of AI agents could do to white-collar knowledge work what assembly lines did to manufacturing.

But the real power of agents comes when they can work as a team.

And it’s not just office work. Multi-agent tools like Google DeepMind’s Co-Scientist let researchers use teams of AI agents to coordinate literature searches, generate and test hypotheses, design experiments, and more.

But the real power of agents comes when they can work as a team. Instead of lone-wolf bots carrying out single tasks, such as using a browser to make a restaurant reservation or sending you a summary of your inbox, new tools can yoke together multiple agents, give each of them a different job, and orchestrate their behaviors so that they all pull together to complete more complex tasks than an individual agent could do by itself.

Think of multi-agent systems as the new assembly lines. Henry Ford’s innovation upended entire industries last century. In theory, networks of AI agents could do to white-collar knowledge work what assembly lines did to manufacturing.

Discovery should focus on trust boundaries, authentication flows, parsers, shared services, and legacy code that still sits on critical paths.

The scariest part of Mythos is not that one lab has a gated model. It is that the core workflow primitives behind representative findings are no longer confined to a single lab's private stack.

The real issue is not whether defenders can get access to another model. It is whether they can turn model capability into something a security team can trust and use every day.

Public models can already spot that a security-relevant check is missing in the right code path, but they can still miss the actual invariant being violated and therefore misstate the impact.

The takeaway is not whether Mythos is better or more powerful. It is that public models can already achieve much the same results.

If public models can already do useful work inside that kind of workflow, then the story is not 'Anthropic has a magical cyber artifact.' The story is that serious AI-assisted vulnerability research is no longer confined to a single frontier lab.

The real challenge is validating outputs, prioritizing what matters, and operationalizing them.

What happens is that weak models hallucinate (sometimes causally hitting a real problem) that there is a lack of validation of the start of the window... without understanding why they, if put together, create an issue.

So, cyber security of tomorrow will not be like proof of work in the sense of 'more GPU wins'; instead, better models, and faster access to such models, will win.

Stronger models hallucinate less, so they can't see the problem in any side of the spectrum: the hallucination side of small models, and the real understanding side of Mythos.

you can run an inferior model for an infinite number of tokens, and it will never realize(*) that the lack of validation of the start window, if put together with the integer overflow, then put together with the fact the branch where the node should never be NULL is entered regardless, will produce the bug.

Keeping a human in the loop may not provide the safeguard people imagine, because the human cannot know the AI's intention before it acts.

Huge advances have been made in developing and building more capable models, driven by record investments—forecast by Gartner to grow to around $2.5 trillion in 2026 alone. In contrast, the investment in understanding how the technology works has been minuscule.

The immediate danger is not that machines will act without human oversight; it is that human overseers have no idea what the machines are actually 'thinking.'

Including design intent in Claude Code handoffs has made the jump from prototype to production seamless.

Claude packages everything into a handoff bundle that you can pass to Claude Code with a single instruction.

Founders and Account Executives can go from a rough outline to a complete, on-brand deck in minutes

What used to take a week of back-and-forth between briefs, mockups, and review rounds now happens in a single conversation.

Our most complex pages, which took 20+ prompts to recreate in other tools, only required 2 prompts in Claude Design.

Claude keeps its responses focused and concise so as to avoid potentially overwhelming the user with overly-long responses

Claude 4.6 had a section specifically clarifying that 'Donald Trump is the current president of the United States and was inaugurated on January 20, 2025'

If people ask Claude to give a simple yes or no answer... Claude can decline to offer the short response

Claude calls tool_search to check whether a relevant tool is available but deferred

the person typically wants Claude to make a reasonable attempt now, not to be interviewed first

Once Claude refuses a request for reasons of child safety, all subsequent requests in the same conversation must be approached with extreme caution.

the move from pattern matching to understanding cause and effect

LLMs actually work under the hood

the ability to keep learning after training and the move from pattern matching to understanding cause and effect

transformers update their predictions in a precise, mathematically predictable way as they process new information

Research has shown that involving workers' perspectives in the design of workplace technologies promotes sustainable improvements in productivity and well-being.

A central pattern emerging in generative AI is a shift from 'thinking by doing' (e.g. writing a document) toward 'choosing from outputs' (e.g. prompting AI to write a document).

LLMs take knowledge from millions of people who have written web content or posted in places like Reddit and Wikipedia, interacted with chatbots, and generated other types of data, and make that available to individuals on demand.

In one U.S. survey, 40% of employees said they had received 'workslop', i.e. AI-generated content that looks polished but isn't accurate or useful, in the past month.

Entry-level roles rely less on experience and knowledge and are easier to automate. Empirical evidence suggests employment for workers aged 22–25 in highly AI-exposed jobs declined by 16% relative to similar but less-exposed roles

Cursor still uses and sells access to Claude and GPT models even as both firms roll out their own coding tools, an awkward arrangement that this new SpaceX partnership may be designed to eventually escape.

Either figure would represent a significant expense for SpaceX, which is widely seen to be losing money following the acquisition of xAI and the social media network X and is planning extensive capital investment.

The deal won’t shock those who follow the industry closely. Last week, it was reported that xAI would begin renting computing power from its data centers to Cursor, with the coding startup using tens of thousands of xAI chips to train its latest AI model.

Neither Cursor nor xAI has proprietary models that can match the leading offerings from Anthropic and OpenAI — the same companies now competing directly with Cursor for the developer market.

Members have been using Mythos regularly since gaining access — providing screenshots and a live demonstration of the model as evidence to _Bloomberg_ — though reportedly not for cybersecurity purposes in an attempt to avoid detection by Anthropic.

The group accessed Mythos by using knowledge of Anthropic’s other model formats obtained from a recent [Mercor data breach](https://www.theverge.com/ai-artificial-intelligence/907083/a-company-that-makes-ai-training-data-has-been-hit-by-a-security-breach) to make “an educated guess” about its online location.

Official access to the model is limited to a handful of companies through the [Project Glasswing initiative](https://www.theverge.com/ai-artificial-intelligence/908114/anthropic-project-glasswing-cybersecurity), including Nvidia, Google, Amazon Web Services, Apple, and Microsoft.

Anthropic currently has no plans to release the model publicly due to concerns that it could be weaponized.

It also surpasses all peer-scale dense models by a wide margin.

With only 27B parameters, it outperforms the Qwen3.5-397B-A17B (397B total / 17B active) on every major coding benchmark

It also surpasses all peer-scale dense models by a wide margin.

TPU 8i is designed with more memory bandwidth to serve the most latency-sensitive inference workloads, which is critical because interactions between agents at scale magnify even small inefficiencies.

Both chips can run various workloads, but specialization unlocks significant efficiencies and gains.

By customizing and co-designing silicon with hardware, networking and software, including model architecture and application requirements, we can deliver dramatically more power efficiency and absolute performance.

These two chips are designed to power our custom-built supercomputers, to drive everything from cutting-edge model training and agent development, to massive inference workloads.

TypeScript 7.0 now performs many steps in parallel, including parsing, type-checking, and emitting.

We have reworked our JavaScript support to be more consistent with how we analyze TypeScript files.

The stable release of TypeScript 7.0 will be published under the `typescript` package and will use the `tsc` entry point.

The new Go codebase was methodically ported from our existing implementation rather than rewritten from scratch.

We spent days loading the system with hundreds of threads, refining rough edges and polishing corners that developers may never see.

Multi-agent orchestration isn't new, but we believe we've built a great experience for working with agents at scale.

What we've found works best for crafting high-quality software is somewhere in between: using AI, and also engaging directly with code.

At one extreme, there's [fully giving into the vibes], and at the other extreme, there's [disabling all AI features].

Ask ten different programmers how they use AI, and you can get ten different answers.

All imagine that in the not-too-distant future many of us will designate some tasks that we currently undertake with our own brains and fingers on a physical PC to an agent that uses a virtual PC.

Meta is not alone in pursuing such a vision: Anthropic debuted tech capable of doing this [in 2024] and OpenAI last year announced [“Operator”] – a tool that can use a web browser on a human’s behalf.

Meta feels AI models don’t understand how people use computers, so the company needs real-life examples of how meatbags click their way through a working day so it can build agents.

Meta, the company built on watching everything its billions of users do online so it can keep them clicking on ragebait and targeted ads, is reportedly now installing surveillance software on employees’ work computers.

The fact that the RL model has larger improvements on Levenshtein Distance and Added Cognitive Complexity than on Pass@1 is further evidence that it is not just memorizing corruption reversals but has actually generalized to minimal editing.

Reasoning models are generally assumed to be better at coding tasks, and they do score higher on Pass@1.

Among the latest frontier models, GPT-5.4 over-edits the most.

While the output passes the tests and is functionally correct, the diff is enormous, and none of those additions were asked for or even necessary.

A common piece of advice for working with AI coding tools is to simply write more tests because if the tests pass, the code is fine.

Because agents have memory and can be guided and corrected in conversation, they get better as teams use them.

Workspace agents can gather context from the right systems, follow team processes, ask for approval when needed, and keep work moving across tools.

They run in the cloud, so they can keep working even when you’re not.

AI has already helped people work faster on their own, but many of the most important workflows inside an organization depend on shared context, handoffs, and decisions across teams.

Workspace agents will be free until May 6, 2026, with credit-based pricing starting on that date.

Admins can also manage who has access to use, build, and share agents.

Because agents have memory and can be guided and corrected in conversation, they get better as teams use them.

They run in the cloud, so they can keep working even when you’re not.

AI has already helped people work faster on their own, but many of the most important workflows inside an organization depend on shared context, handoffs, and decisions across teams.

That matters because AI hype is dying down, and companies are shifting focus from buzzy pilots to deployment and integration, where cheaper and more customizable tools tend to win.

US tech CEOs believe the best models should stay proprietary, partly so they can recoup enormous training costs and partly out of concern that powerful frontier models could be weaponized. Chinese labs, for their part, are not purely idealistic: Open-source is not only free advertising but also a shrewd workaround.

Chinese labs, for their part, are not purely idealistic: Open-source is not only free advertising but also a shrewd workaround. Without access to cutting-edge chips restricted by US export controls, releasing models openly accelerates the cycle of external feedback and contributions that compensates for constrained compute.

Chinese open-weight models accounted for 17.1% of global AI model downloads over the year ending in August 2025. That narrowly surpassed the US share of 15.86%—the first time China had led in this metric.

Telling people to avoid using generative AI is increasingly telling them they must avoid taking part in society.

We have not really begun to make this progress with AI. Why, for example, is this dashboard not found on a government website?

to stand out from the AI-generated pack we will need to become so weird and unexpected as to be off-putting to most people

The AI has learned to code. The AI is building itself.

Is this what we signed up for? Is today the day? Did the drones wake up? Did it achieve consciousness? Is it alive?

We have not really begun to make this progress with AI. Why, for example, is this dashboard not found on a government website?

to stand out from the AI-generated pack we will need to become so weird and unexpected as to be off-putting to most people

The 21st-century average American lies in bed staring at their phone. ... Talking for hours and ages to melted sand.

The future is exciting – perhaps the vision of truly self-serve analytics can be fully realized, and BI, data analytics, and data science can be transformed through AI.

They will have to go through our journey above of ingesting data, collecting tribal knowledge, and more – and they will have to do so for each individual customer they work with.

Many have realized through time in market that the key to effective data agents is actually building the relevant context layer. As a result, some have evolved to encompass data context construction as a key part of their products.

While model capabilities have improved dramatically for use cases like codegen and mathematical reasoning, they still lag behind on the data side (as evidenced through SQL benchmarks like Spider 2.0 and Bird Bench).

The general idea was that a model should be able to take in a natural language query as an initial input, reason over existing data systems, and generate corresponding SQL code in traditional business intelligence (BI) fashion to pull the right data and answer the initial question accordingly.

The benefit of using LLMs is that a lot of the initial context gathering can be done in an automated way. An emphasis of focus should be on high signal context – for example, looking through past query history can be high signal in determining the most referenced tables and most common joins, and data modeling solutions like dbt or LookML can provide clear definitions for business metrics.

A modern data context layer should essentially become a superset of what a semantic layer would traditionally cover. Sure, specific metric definitions can be hard-coded, but a modern context layer should include more to ensure agent autonomy – canonical entities, identity resolution, specific instructions to dissect tribal knowledge, proper governance guidance, and more.

While the initial system has been set up correctly, data systems are never static and as a result the context layer shouldn't be either. Data sources and formats can change upstream and individuals may have custom instructions they'll want to add and modify based on changing business requirements.

This piece will primarily focus on data context that ties together traditional systems of record. An equally important and overlapping opportunity is also capturing an organization's decisions and workflow logic so truly multipurpose agents can be built that are properly grounded in all of an organization's data and decisioning context.

The modern data stack has undergone a decade+ transition from disparate data sources to consolidated data and cleaned definitions (which is good), but even then the consolidation is never perfect and a lot of messiness is introduced.

We are at an interesting point in time of market development, where the problem of a lack of context has become apparent, but we are still in the early innings of building solutions.

The OpenAI team recently published a fantastic piece detailing the creation of their own internal data agent. It's a transparent detail of a very detailed and elegant implementation – but points to the long journey required to get there.

In this way the context layer can become a multi-dimensional corpus where code lives alongside natural language, capturing any context an agent might need.

Some of the most important context is implicit, conditional, and historically contingent, and only exists as tribal knowledge inside teams.

A traditional semantic layer in the context of BI is great for specific metric definitions (like revenue, churn, ARPU). However, they are usually hand constructed by data teams using very specific syntax through a dedicated layer like LookML and are connected directly to a BI tool like Looker.

The crux of the problem at hand is that the agent isn't given the proper business context to answer even the most basic questions. This is representative of a larger gap that's present in building automated AI systems within organizations – there needs to be up-to-date and maintained context that not only understands how an enterprise works and how the data systems are structured, but also maintains the tribal knowledge to tie everything together.

To overcome this blocker, a team member hard codes the exact revenue and timeframe definitions. The data agent continues chugging along but quickly runs into challenge #2 – where are the right data sources? Which ones are the right sources of truth?

Over the past year, the market has realized that data and analytics agents are essentially useless without the right context – they aren't able to tease apart vague questions, decipher business definitions, and reason across disparate data effectively.

它对应的agent能获取你的邮箱权限，它知道你一直在等待一个offer，当你收到打开这个offer后，Mira会理解这种心情，开始开心跳舞和闪灯，与你一起庆祝。

通过AI分析将页面上的可交互元素汇集到鼠标周围，并能根据用户兴趣提供额外功能（

小红书已经是AI创业公司和产品的重要分发渠道、产品试错和运营用户的默认场所（可能没有之一）。

他们想要通过统一的Agent框架来解决这些问题，把这些原始的、非结构化的信号数据也纳入端到端处理的范畴

context management plus engineering improvements may well push the task horizon to weeks or even months.

if a model cannot learn new things while performing a task, it will struggle when the task horizon grows very long.

"The set of efforts aimed at breaking past the feasible horizon of current techniques."

new techniques may initially underperform existing ones but eventually surpass them — a pattern we've seen repeatedly, most recently in the wave of agentic coding progress

We can treat the task horizon that an LLM can reliably handle as a north-star metric for model progress, analogous to transistor density in Moore's Law

The key reason for the confusion is that people think in terms of methods that each contribute a discrete piece to the system — pretraining, SFT, RL.

I'd view continual learning more as an "arrow" than a "line" — it's the collective effort to push the task horizon that an LLM can reliably handle.

settings.local.json、mcpServers headers 里的 Bearer Token 不能明文外带。

包括 47 种数据：agents、skills、hooks、MCP 配置、会话历史、自定义规则……

不做 n×m 点对点迁移，通过 neuDrive 做 n→1→m 中心化枢纽：一次导出，任意 Agent 消费。

多年积累的对话、定制 Agent、项目记忆、MCP 配置、Skill 库——一次风控就可能全部失联。

Arbitrageur: Knows pₜ. Sweeps every resting ask below pₜ and every resting bid above pₜ. Infinite capital, never rests orders.

You start with $1,000 cash, 0 YES, and 0 NO. Minting one YES and one NO costs $1.

Competitor: Static hidden-liquidity ladder. Quotes every tick outside its spread with fixed notional. Refills consumed levels at a fixed offset next step. Never re-centers.

With ~2 expected jumps per simulation at default intensity, each jump is a significant information shock.

The diffusion term is fixed across all simulations. The regime-level variation comes entirely from the jump parameters - intensity, mean, and variance - which are randomized per simulation.

You quote before the next price move, so you are always exposed to adverse selection.

Retail fills generate positive edge (you captured the spread). Arb fills generate negative edge (the arbitrageur took stale quotes).

Your advantage comes from adapting to market conditions it ignores.

Configuration is managed via environment variables. See src/aegis_core/config.py for all available settings.

Infrastructure Provisioning cd deploy/terraform/aliyun terraform init terraform plan terraform apply Helm Deployment cd deploy/helm helm install aegis-core ./aegis-core \ --namespace aegis \ --create-namespace \ --set image.repository=<acr-registry>/aegis-core \ --set image.tag=lat

Quick Start # Clone the repository git clone https://github.com/fxp/aegis-core.git cd aegis-core # Start all services with Docker Compose docker-compose up -d # The API is available at http://localhost:8000 # Health check: http://localhost:8000/health

Unified interface for interacting with different LLM providers (Claude, OpenAI, local models via vLLM/Ollama). Includes tool definitions for security operations (shell, file I/O, network, debugger) and cost/token tracking.

Tracks the evolution of LLM security capabilities across benchmarks (CyberGym, Cybench, etc.), calculates capability doubling times, detects emergence patterns, and monitors cost-efficiency trends.

Real-time monitoring of agent actions with a 12-category anomaly detection system derived from frontier model safety evaluations. Three-level alert system: PROHIBITED (immediate block), HIGH_RISK_DUAL_USE (human review), DUAL_USE (log and track).

Aegis Core provides the foundational infrastructure for orchestrating LLM-based security agents, monitoring their behavior, and tracking the evolution of AI security capabilities over time.

That includes ongoing partnerships with national laboratories such as Los Alamos National Laboratory, where we are exploring AI-guided protein and catalyst design, including the ability of AI systems to modify biological structures while preserving or improving key functional properties. Over time, we expect these systems to become increasingly capable partners in discovery—helping scientists move faster from question to evidence, from evidence to insight, and from insight to new treatments for patients.

The Life Sciences model was developed with heightened enterprise-grade security controls and strengthened access management, enabling professional scientific use in governed research environments.

They provide access to more than 50 public multi-omics databases, literature sources, and biology tools, and offer a flexible starting point for common repeatable workflows.

helping scientists move faster from question to evidence, from evidence to insight, and from insight to new treatments for patients.

We will continue improving the model's biological reasoning, expanding support for tool-heavy and long-horizon research workflows, and working closely with leading scientific institutions to evaluate real-world impact.

Over time, these systems could help life sciences organizations discover breakthroughs that wouldn't otherwise be possible, with a much higher rate of success.

Scientists must work across large volumes of literature, specialized databases, experimental data, and evolving hypotheses in order to generate and evaluate new ideas.

The model is named after Rosalind Franklin, whose rigorous research helped reveal the structure of DNA and laid foundations for modern molecular biology.

Performance was compared against 57 historical scores from human experts in the AI-bio field.

GPT‑Rosalind is now available as a research preview in ChatGPT, Codex, and the API for qualified customers through our trusted access program.

Organic chemistry Protein understanding Genomics Experimental design and analysis Tool usage

These skills act as an orchestration layer that helps scientists work through broad, ambiguous, and multi-step questions more effectively.

The most notable improvement comes from CloningQA, which requires end-to-end design of DNA and enzyme reagents for molecular cloning protocols.

When evaluated directly in the Codex app, best-of-ten model submissions ranked above the 95th percentile of human experts on the prediction task and around the 84th percentile of human experts on the sequence generation task.

Progress in the life sciences is constrained not only by the difficulty of the underlying science, but by the complexity of the research workflows themselves.

On average, it takes roughly 10 to 15 years to go from target discovery to regulatory approval for a new drug in the United States.

In our own testing, the net effect is favorable—token usage across all effort levels is improved on an internal coding evaluation, as shown below—but we recommend measuring the difference on real traffic.

Claude Opus 4.7 demonstrates strong substantive accuracy on BigLaw Bench for Harvey, scoring 90.9% at high effort with better reasoning calibration on review tables and noticeably smarter handling of ambiguous document editing tasks.

Claude Opus 4.7 is a meaningful step up for Warp. Opus 4.6 is one of the best models out there for developers, and this model is measurably more thorough on top of that. It passed Terminal Bench tasks that prior Claude models had failed

Opus 4.7 uses an updated tokenizer that improves how the model processes text. The tradeoff is that the same input can map to more tokens—roughly 1.0–1.35× depending on the content type.

For Ramp, Claude Opus 4.7 stands out in agent-team workflows. We're seeing stronger role fidelity, instruction-following, coordination, and complex reasoning, especially on engineering tasks that span tools, codebases, and debugging context.

Claude Opus 4.7 feels like a real step up in intelligence. Code quality is noticeably improved, it's cutting out the meaningless wrapper functions and fallback scaffolding that used to pile up, and fixes its own code as it goes.

Claude Opus 4.7 is the most capable model we've tested at Quantium. Evaluated against leading AI models through our proprietary benchmarking solution, the biggest gains showed up where they matter most: reasoning depth, structured problem-framing, and complex technical work.

Claude Opus 4.7 is a solid upgrade with no regressions for Vercel. It's phenomenal on one-shot coding tasks, more correct and complete than Opus 4.6, and noticeably more honest about its own limits.

Opus 4.7 is better at using file system-based memory. It remembers important notes across long, multi-session work, and uses them to move on to new tasks that, as a result, need less up-front context.

Opus 4.7 introduces a new `xhigh` ('extra high') effort level between `high` and `max`, giving users finer control over the tradeoff between reasoning and latency on hard problems.

Claude Opus 4.7 is measurably better than Opus 4.6 for Bolt's longer-running app-building work, up to 10% better in the best cases, without the regressions we've come to expect from very agentic models.

Claude Opus 4.7 passed three TBench tasks that prior Claude models couldn't, and it's landing fixes our previous best model missed, including a race condition.

For the computer-use work that sits at the heart of XBOW's autonomous penetration testing, the new Claude Opus 4.7 is a step change: 98.5% on our visual-acuity benchmark versus 54.5% for Opus 4.6.

Claude Opus 4.7 is the best model in the world for building dashboards and data-rich interfaces. The design taste is genuinely surprising—it makes choices I'd actually ship.

For complex multi-step workflows, Claude Opus 4.7 is a clear step up: plus 14% over Opus 4.6 at fewer tokens and a third of the tool errors. It's the first model to pass our implicit-need tests.

Claude Opus 4.7 autonomously built a complete Rust text-to-speech engine from scratch—neural model, SIMD kernels, browser demo—then fed its own output through a speech recognizer to verify it matched the Python reference.