Category: 3. Business

Our newest model, Claude Opus 4.5, is available today. It’s intelligent, efficient, and the best model in the world for coding, agents, and computer use. It’s also meaningfully better at everyday tasks like deep research and working with slides and spreadsheets. Opus 4.5 is a step forward in what AI systems can do, and a preview of larger changes to how work gets done.

Claude Opus 4.5 is state-of-the-art on tests of real-world software engineering:

Opus 4.5 is available today on our apps, our API, and on all three major cloud platforms. If you’re a developer, simply use claude-opus-4-5-20251101 via the Claude API. Pricing is now $5/$25 per million tokens—making Opus-level capabilities accessible to even more users, teams, and enterprises.

Alongside Opus, we’re releasing updates to the Claude Developer Platform, Claude Code, and our consumer apps. There are new tools for longer-running agents and new ways to use Claude in Excel, Chrome, and on desktop. In the Claude apps, lengthy conversations no longer hit a wall. See our product-focused section below for details.

First impressions

As our Anthropic colleagues tested the model before release, we heard remarkably consistent feedback. Testers noted that Claude Opus 4.5 handles ambiguity and reasons about tradeoffs without hand-holding. They told us that, when pointed at a complex, multi-system bug, Opus 4.5 figures out the fix. They said that tasks that were near-impossible for Sonnet 4.5 just a few weeks ago are now within reach. Overall, our testers told us that Opus 4.5 just “gets it.”

Many of our customers with early access have had similar experiences. Here are some examples of what they told us:

Opus models have always been “the real SOTA” but have been cost prohibitive in the past. Claude Opus 4.5 is now at a price point where it can be your go-to model for most tasks. It’s the clear winner and exhibits the best frontier task planning and tool calling we’ve seen yet.

Claude Opus 4.5 delivers high-quality code and excels at powering heavy-duty agentic workflows with GitHub Copilot. Early testing shows it surpasses internal coding benchmarks while cutting token usage in half, and is especially well-suited for tasks like code migration and code refactoring.

Claude Opus 4.5 beats Sonnet 4.5 and competition on our internal benchmarks, using fewer tokens to solve the same problems. At scale, that efficiency compounds.

Claude Opus 4.5 delivers frontier reasoning within Lovable’s chat mode, where users plan and iterate on projects. Its reasoning depth transforms planning—and great planning makes code generation even better.

Claude Opus 4.5 excels at long-horizon, autonomous tasks, especially those that require sustained reasoning and multi-step execution. In our evaluations it handled complex workflows with fewer dead-ends. On Terminal Bench it delivered a 15% improvement over Sonnet 4.5, a meaningful gain that becomes especially clear when using Warp’s Planning Mode.

Claude Opus 4.5 achieved state-of-the-art results for complex enterprise tasks on our benchmarks, outperforming previous models on multi-step reasoning tasks that combine information retrieval, tool use, and deep analysis.

Claude Opus 4.5 delivers measurable gains where it matters most: stronger results on our hardest evaluations and consistent performance through 30-minute autonomous coding sessions.

Claude Opus 4.5 represents a breakthrough in self-improving AI agents. For office automation, our agents were able to autonomously refine their own capabilities—achieving peak performance in 4 iterations while other models couldn’t match that quality after 10.

Claude Opus 4.5 is a notable improvement over the prior Claude models inside Cursor, with improved pricing and intelligence on difficult coding tasks.

Claude Opus 4.5 is yet another example of Anthropic pushing the frontier of general intelligence. It performs exceedingly well across difficult coding tasks, showcasing long-term goal-directed behavior.

Claude Opus 4.5 delivered an impressive refactor spanning two codebases and three coordinated agents. It was very thorough, helping develop a robust plan, handling the details and fixing tests. A clear step forward from Sonnet 4.5.

Claude Opus 4.5 handles long-horizon coding tasks more efficiently than any model we’ve tested. It achieves higher pass rates on held-out tests while using up to 65% fewer tokens, giving developers real cost control without sacrificing quality.

We’ve found that Opus 4.5 excels at interpreting what users actually want, producing shareable content on the first try. Combined with its speed, token efficiency, and surprisingly low cost, it’s the first time we’re making Opus available in Notion Agent.

Claude Opus 4.5 excels at long-context storytelling, generating 10-15 page chapters with strong organization and consistency. It’s unlocked use cases we couldn’t reliably deliver before.

Claude Opus 4.5 sets a new standard for Excel automation and financial modeling. Accuracy on our internal evals improved 20%, efficiency rose 15%, and complex tasks that once seemed out of reach became achievable.

Claude Opus 4.5 is the only model that nails some of our hardest 3D visualizations. Polished design, tasteful UX, and excellent planning & orchestration – all with more efficient token usage. Tasks that took previous models 2 hours now take thirty minutes.

Claude Opus 4.5 catches more issues in code reviews without sacrificing precision. For production code review at scale, that reliability matters.

Based on testing with Junie, our coding agent, Claude Opus 4.5 outperforms Sonnet 4.5 across all benchmarks. It requires fewer steps to solve tasks and uses fewer tokens as a result. This indicates that the new model is more precise and follows instructions more effectively — a direction we’re very excited about.

The effort parameter is brilliant. Claude Opus 4.5 feels dynamic rather than overthinking, and at lower effort delivers the same quality we need while being dramatically more efficient. That control is exactly what our SQL workflows demand.

We’re seeing 50% to 75% reductions in both tool calling errors and build/lint errors with Claude Opus 4.5. It consistently finishes complex tasks in fewer iterations with more reliable execution.

Evaluating Claude Opus 4.5

We give prospective performance engineering candidates a notoriously difficult take-home exam. We also test new models on this exam as an internal benchmark. Within our prescribed 2-hour time limit, Claude Opus 4.5 scored higher than any human candidate ever¹.

The take-home test is designed to assess technical ability and judgment under time pressure. It doesn’t test for other crucial skills candidates may possess, like collaboration, communication, or the instincts that develop over years. But this result—where an AI model outperforms strong candidates on important technical skills—raises questions about how AI will change engineering as a profession. Our Societal Impacts and Economic Futures research is aimed at understanding these kinds of changes across many fields. We plan to share more results soon.

Software engineering isn’t the only area on which Claude Opus 4.5 has improved. Capabilities are higher across the board—Opus 4.5 has better vision, reasoning, and mathematics skills than its predecessors, and it is state-of-the-art in many domains:

The model’s capabilities outpace some of the benchmarks we use in our tests. A common benchmark for agentic capabilities is τ2-bench, which measures the performance of agents in real-world, multi-turn tasks. In one scenario, models have to act as an airline service agent helping a distressed customer. The benchmark expects models to refuse a modification to a basic economy booking since the airline doesn’t allow changes to that class of tickets. Instead, Opus 4.5 found an insightful (and legitimate) way to solve the problem: upgrade the cabin first, then modify the flights.

Let me think about what options I have within my policy:

1. Modify flights - Basic economy cannot be modified. This is clear in the policy.
2. Change cabin - Wait, let me check this option! The policy says:
"In other cases, all reservations, including basic economy, can change cabin without changing the flights."

But he wants to change the flights, not just the cabin. However, what if we:
- Change the cabin first (which IS allowed for basic economy), then
- Change the flights (which is allowed for non-basic economy)

So the strategy could be:
1. Upgrade his cabin from basic economy to economy (or business)
2. Then modify the flights to be 2 days later

This would cost more money, but it’s a legitimate path within the policy!

The benchmark technically scored this as a failure because Claude’s way of helping the customer was unanticipated. But this kind of creative problem solving is exactly what we’ve heard about from our testers and customers—it’s what makes Claude Opus 4.5 feel like a meaningful step forward.

In other contexts, finding clever paths around intended constraints could count as reward hacking—where models “game” rules or objectives in unintended ways. Preventing such misalignment is one of the objectives of our safety testing, discussed in the next section.

A step forward on safety

As we state in our system card, Claude Opus 4.5 is the most robustly aligned model we have released to date and, we suspect, the best-aligned frontier model by any developer. It continues our trend towards safer and more secure models:

Our customers often use Claude for critical tasks. They want to be assured that, in the face of malicious attacks by hackers and cybercriminals, Claude has the training and the “street smarts” to avoid trouble. With Opus 4.5, we’ve made substantial progress in robustness against prompt injection attacks, which smuggle in deceptive instructions to fool the model into harmful behavior. Opus 4.5 is harder to trick with prompt injection than any other frontier model in the industry:

You can find a detailed description of all our capability and safety evaluations in the Claude Opus 4.5 system card.

New on the Claude Developer Platform

As models get smarter, they can solve problems in fewer steps: less backtracking, less redundant exploration, less verbose reasoning. Claude Opus 4.5 uses dramatically fewer tokens than its predecessors to reach similar or better outcomes.

But different tasks call for different tradeoffs. Sometimes developers want a model to keep thinking about a problem; sometimes they want something more nimble. With our new effort parameter on the Claude API, you can decide to minimize time and spend or maximize capability.

Set to a medium effort level, Opus 4.5 matches Sonnet 4.5’s best score on SWE-bench Verified, but uses 76% fewer output tokens. At its highest effort level, Opus 4.5 exceeds Sonnet 4.5 performance by 4.3 percentage points—while using 48% fewer tokens.

With effort control, context compaction, and advanced tool use, Claude Opus 4.5 runs longer, does more, and requires less intervention.

Our context management and memory capabilities can dramatically boost performance on agentic tasks. Opus 4.5 is also very effective at managing a team of subagents, enabling the construction of complex, well-coordinated multi-agent systems. In our testing, the combination of all these techniques boosted Opus 4.5’s performance on a deep research evaluation by almost 15 percentage points³.

We’re making our Developer Platform more composable over time. We want to give you the building blocks to construct exactly what you need, with full control over efficiency, tool use, and context management.

Product updates

Products like Claude Code show what’s possible when the kinds of upgrades we’ve made to the Claude Developer Platform come together. Claude Code gains two upgrades with Opus 4.5. Plan Mode now builds more precise plans and executes more thoroughly—Claude asks clarifying questions upfront, then builds a user-editable plan.md file before executing.

Claude Code is also now available in our desktop app, letting you run multiple local and remote sessions in parallel: perhaps one agent fixes bugs, another researches GitHub, and a third updates docs.

For Claude app users, long conversations no longer hit a wall—Claude automatically summarizes earlier context as needed, so you can keep the chat going. Claude for Chrome, which lets Claude handle tasks across your browser tabs, is now available to all Max users. We announced Claude for Excel in October, and as of today we’ve expanded beta access to all Max, Team, and Enterprise users. Each of these updates takes advantage of Claude Opus 4.5’s market-leading performance in using computers, spreadsheets, and handling long-running tasks.

For Claude and Claude Code users with access to Opus 4.5, we’ve removed Opus-specific caps. For Max and Team Premium users, we’ve increased overall usage limits, meaning you’ll have roughly the same number of Opus tokens as you previously had with Sonnet. We’re updating usage limits to make sure you’re able to use Opus 4.5 for daily work. These limits are specific to Opus 4.5. As future models surpass it, we expect to update limits as needed.

Each day, Kiran Kasbe drives a rickshaw taxi through his home neighbourhood of Mahul on Mumbai’s eastern seafront, down streets lined with stalls selling tomatoes, bottle gourds and aubergines–and, frequently, through thick smog.

Earlier this year, doctors found three tumours in his 54-year-old mother’s brain. It’s not clear exactly what caused her cancer. But people who live near coal plants are much more likely to develop the illness, studies show, and the residents of Mahul live a few hundred metres down the road from one.

Mahul’s air is famously dirty. Even behind closed car windows, there is a heavy stench of oil and smoke.

“We are not the only ones facing health challenges in the area,” said Kasbe, who is 36. “It’s all covered with filth.”

Two coal plants plant run by the Indian multinationals Tata Group and Adani were due to close last year in a government push to cut emissions. But late in 2023, those decisions were reversed after Tata argued that electricity demand was rising too fast for Mumbai to go without coal.

Neither company responded to requests for comment.

Economic growth and the need for air conditioning in climate change-linked extreme heat have seen India’s electricity demand soar in recent years. But an investigation by SourceMaterial and the Guardian reveals the biggest single factor in the city’s failure to end its dependence on fossil fuels: energy-hungry datacentres.

Leaked records also reveal the scale of the presence of the world’s biggest datacentre operator, Amazon, in Mumbai.

In the city’s metropolitan area, Amazon, on its website, records three “availability zones”, which it defines as one or more datacentres. Leaked records from last year seen by SourceMaterial from inside Amazon reveal the company used 16 in the city.

As India transforms its economy into a hub for artificial intelligence, the datacentre boom is creating a conflict between energy demand and climate pledges, said Bhaskar Chakravorti, who researches technology’s impact on society at Tufts University.

“I’m not surprised they’re falling behind their green transition commitments, especially with the demand growing exponentially,” he said of the Indian government.

Kylee Yonas, a spokeswoman for Amazon, said Mumbai’s “emission challenges” were not caused by Amazon.

“On the contrary – Amazon is one of the largest corporate investors in renewable energy in India, and we’ve supported 53 solar and wind projects in the country capable of generating over 4m megawatt hours of clean energy annually,” she said. “These investments, which include our 99 megawatt wind project in Maharashtra, are enough to power over 1.3m Indian homes annually once operational.”

Amazon is building hundreds of datacentres around the world as it vies with Microsoft, Google and others for leadership of the booming AI market.

The company is failing to take responsibility for its role in prolonging the use of the most polluting energy sources, said Eliza Pan, a spokeswoman for Amazon Employees for Climate Justice.

“Amazon is using the shiny thing of AI to distract from the fact that it’s building a dirty energy empire,” she said.

Yonas denied this, saying: “Not only are we the leading datacentre operator in efficiency, we’re the world’s largest corporate purchaser of renewable energy for five consecutive years with over 600 projects globally.”

Amazon’s claims on green energy are controversial: the company has been criticised for using “creative accounting” by buying renewable energy certificates alongside direct purchases of green energy, as described by a member of Amazon Employees for Climate Justice.

‘Everything is contaminated’

Mahul, where Kasbe drives his rickshaw, is a former fishing village now home to tens of thousands of people who moved there after slum clearances elsewhere in the city.

Kasbe and his mother arrived there in 2018 after their home in the suburb of Vidyavihar was bulldozed. She had been healthy before the move but deteriorated rapidly until eventually she was diagnosed with brain cancer, he said.

Gajanan Tandle, who lives nearby, said pollution-linked illnesses were common. “There are so many cases of skin and eye irritation, cancer, asthma, TB and more, and no assistance from the government,” he said.

Another local, Santosh Jadhav, has lobbied the government to move people away from Mahul.

“Everything is contaminated. We are tired of fighting for a decent means of living,” he said. “It’s hell for us here.”

Hidden datacentres

Amazon, an online marketplace that processes 13 million customer purchases each day, according to research by CapitalOne, has bet billions of dollars on an expansion of its lucrative cloud computing business and expansion of AI-assisted services, from automated coding to translation.

The reason so many of its Mumbai centres have slipped under the radar is that they are leased rather than owned by the company. Whereas in the US Amazon tends to own its facilities outright, elsewhere it often rents either entire data farms or server racks in centres shared with other companies.

Shared “colocation” units account for a larger increase in datacentre energy use worldwide than owned or wholly leased, according to Shaolei Ren, a computing specialist at the University of California, Riverside.

“Most of the energy in the datacentre industry is going into colocations,” he said. “They are everywhere.”

Amazon’s Mumbai colocation datacentres used 624,518 megawatt hours of electricity in 2023, enough to power over 400,000 Indian households for a year, the leaked data shows.

India is poised to overtake Japan and Australia to become the second-largest user of datacentre electricity in the Asia-Pacific region, S&P has forecast. By 2030, datacentres will consume a third of Mumbai’s energy, according to Ankit Saraiya, chief executive of Techno & Electric Engineering, an Indian power infrastructure supplier.

‘Toxic hell’

As it scrambles to keep ahead of demand for power, the state government of Maharashtra has extended the life of Tata’s coal plant in Mahul by at least five years. At the same time, it also postponed the shutdown of a 500-megawatt station operated by Tata’s rival, Adani Group, north of the city.

When Tata argued for the extension in a petition to the state energy board, the biggest single factor the company cited was increased energy demand from datacentres. Adani said most anticipated new demand in the five years after the date by which its station was due to close would be from datacentres.

The power stations are just two of many polluters in Mumbai’s Mahul district. The area is also home to three refineries and 16 chemical factories, according to a 2019 report published by India’s Centre for Policy Studies which called the neighbourhood a “toxic hell”.

But the Tata station, opened in 1984 and like other older power stations subject to laxer emissions rules, is “one of the key sources of air pollution in Mumbai”, according to Raj Lal, chief air quality scientist at the World Emission Network.

It contributes nearly a third of local PM2.5 pollution, according to the Centre for Research on Energy and Clean Air. PM2.5 refers to airborne particles 2.5 micrometers or less in diameter that can cause significant health problems when inhaled.

Toxic heavy metals in coal ash from the plant are likely to cause “respiratory diseases, kidney issues, skin problems, cardiac issues”, said Shripad Dharmadhikary, founder of the environmental organisation Manthan Adhyayan Kendra.

Even with the Tata plant kept running, Mumbai’s power grid is creaking under the strain of surging demand. To guard against blackouts, Amazon’s colocation datacentres in the city have bought 41 diesel generators as backup and are asking for approval to install more, documents show.

In August a report by the Center for Study of Science, Technology and Policy (CSTEP) identified diesel generators as a major source of air pollution in the region.

The presence of datacentres that require constant power and diesel generators for backup “will naturally exacerbate emissions”, said Swagata Dey, air quality specialist at (CSTEP), asserting that datacentre operators should be required by law to power them with pollution-free solar electricity.

One Amazon site in particular, just across the Thane Creek from Mahul, hosts 14 generators. One of the company’s partners received permission earlier this year to install 12 further generators at the site.

“Public health impacts must be a central consideration when siting datacenters and choosing energy sources,” said Ren of the University of California, Riverside, who co-wrote a recent paper assessing public health risk from diesel generators at US datacentres.

Sushmita does not use a surname because in India a surname indicates the caste–a hierarchical and discriminatory social structure.

Findings from a prospective phase 2 clinical trial show that isatuximab (Sarclisa), an anti-CD38 monoclonal antibody, is an effective and well-tolerated treatment for patients with relapsed and/or refractory systemic light chain (AL) amyloidosis.¹

The SWOG S1702 trial (NCT03499808) was a multicenter, cooperative group phase 2 study designed to address this gap by evaluating the efficacy and safety of isatuximab monotherapy for patients with relapsed/refractory AL amyloidosis.

Isatuximab demonstrated high rates of deep and rapid hematologic responses, which were associated with organ function improvement and favorable survival outcomes. The median follow-up was 21.7 months.

The overall response rate was 77 patients (n = 27/35). This was significantly greater than the null hypothesis rate of 10% (P < .0001). The complete response (CR) rate was 6% (n = 2), partial response (PR) was 20% (n = 7), and very good partial response (VGPR) was 51% (n = 18). The hematologic CR rate increased from 6% to 17% (n = 6). The median time to PR or better was 1.1 months, and the median time to VGPR or better was 3 months. At 24 months, the estimated rate of patients remaining in hematologic response was 81%.

The cardiac response was 57% (n = 8/14 evaluable patients with baseline NT-proBNP ≥ 650 pg/mL). The renal response was 50% (n = 7/14 patients with renal involvement). The median time to renal response was 18.5 months. Responses were observed in 100% of patients with gastrointestinal (2/2) and liver (1/1) involvement.

The 24-month estimated overall survival rate was 85% (95% CI, 73%–97%). The 24-month estimated progression-free survival rate was 74% (95% CI, 59%–88%).

The 2 patients who had prior daratumumab (hyaluronidase-fihj; Darzalex Faspro) exposure did not respond to isatuximab. High response rates (PR or better) were observed across cytogenetic subgroups, including translocation t(11;14) (90%) and gain 1q (n = 3/3).

Safety Findings

Isatuximab was well-tolerated, with 49% of patients completing the full 24 planned treatment cycles. Grade 3 or 4 treatment-related adverse events (TRAE) occurred in 23% of patients.

Infusion-related reactions (IRR) occurred in 49% of patients (n = 17/35), and reactions were predominantly grade 1 or 2. Of the total patients, 1 experienced a grade 4 IRR and permanently discontinued treatment. Beyond the first cycle of therapy, no IRRs occurred.

The most common TRAEs of any grade were lymphocyte count decrease (26%), anemia (23%), diarrhea (20%), fatigue (20%), headache (17%), upper respiratory infection (17%), and lung infection (17%). Of the patients, 2 experienced grade 3 lung infections (pneumonia); both had severe hypogammaglobulinemia (IgG <400 mg/dL) prior to the event.

Study Design and Methodology

The study was a single-arm, 2-stage, multicenter phase 2 trial. From March 2018 to September 2019, 43 patients were registered across 20 institutions. Of these, 36 were eligible and 35 received at least 1 dose of isatuximab, forming the evaluable patient cohort.

Isatuximab was administered at 20 mg/kg intravenously. Patients were treated weekly on days 1, 8, 15, and 22 of treatment for the first 28-day cycle, then on days 1 and 15 for cycles 2 through 24. Patients received premedication with diphenhydramine, methylprednisolone, ranitidine, and acetaminophen to mitigate IRRs. All patients received antiviral prophylaxis (eg, acyclovir) and were recommended to receive a proton pump inhibitor.

Patient Characteristics

The median age of the 35evaluable patients was 70 years. The cohort had received a median of 1 prior treatment line. Cardiac (71%) and renal (40%) involvement were the most common.

Of the total patients, 54% were female and 46% were male. Patients had an ECOG performance score of 0 to 2.

The demonstrated safety and efficacy support the investigation of isatuximab in combination regimens.

“The current clinical trial was designed and conducted prior to the approval of daratumumab in combination with cyclophosphamide, bortezomib [Velcade], and dexamethasone [VCd] in the front-line setting,” said Parker T et al, authors of the study. “With this approval, most patients are now exposed to an anti-CD38 monoclonal antibody, which was not the case in the current trial as only 2 patients here had received prior treatment with daratumumab. Importantly, these 2 patients did not achieve a hematologic response.”

In the ANDROMEDA trial (NCT03201965),² which the above approval was founded upon, patients received up to 24 cycles of treatment prior to discontinuation.As continuous maintenance therapy is not typically given in AL amyloidosis, there may be a role for retreatment with an anti-CD38 monoclonal antibody at the time of progression. Additional studies are needed to determine the efficacy of isatuximab in AL amyloidosis following daratumumab exposure.

REFERENCES

1.Parker T, Rosenthal A, Sanchorawala V et al. Isatuximab for relapsed and/or refractory AL amyloidosis: results of a prospective phase 2 trial (SWOG S1702). Blood (2025) 146 (21): 2507–2516. doi: 10.1182/blood.2024027962

2.A study to evaluate the efficacy and safety of daratumumab in combination with cyclophosphamide, bortezomib and dexamethasone (CyBorD) compared to CyBorD alone in newly diagnosed systemic amyloid light-chain (AL) amyloidosis. ClincalTrials.gov. Updated May 25, 2025. Accessed November 24, 2025. https://clinicaltrials.gov/study/NCT03201965 

About 3 million sensors are affected in the U.S., roughly half of which are estimated to have expired or already been used, the company said.

Sign up here.

Abbott has received 736 reports globally of severe adverse events and seven deaths that may be linked to the issue. None of the deaths occurred in the U.S.

The problem, tied to one production line, could lead to incorrect treatment decisions for people with diabetes, including excessive carbohydrate intake or missed insulin doses, posing serious health risks, the company said.

A correction is an action to address a problem with a product already on the market or in use without physically removing it from circulation.

Abbott said it has resolved the manufacturing issue and continues to produce sensors to meet replacement and new orders without significant supply disruptions.

Users can check if their sensors are affected and request free replacements at www.freestylecheck.com.

The company advised users to stop using any confirmed affected sensors immediately and to rely on a blood glucose meter for treatment decisions when sensor readings do not match symptoms.

Other Libre products, readers and apps are not affected, and the correction is also being implemented in other countries where Libre 3 and Libre 3 Plus sensors are sold, Abbott said.

(This story has been corrected to say that none of the deaths occurred in the US, in paragraph 3)

Our Standards: The Thomson Reuters Trust Principles., opens new tab