Fable 5.1 and Mythos 5.1 were suddenly launched overnight.
The previous generation of Fable 5 had almost become synonymous with SOTA models, and before its aura had faded, Anthropic had already begun to surpass itself.
Two models were released simultaneously, with Fable 5.1 targeting general users and enterprises, and Mythos 5.1 opening up the same cutting-edge capabilities to "verified cybersecurity and life sciences institutions".
In terms of pricing, the base price remains unchanged, with Fable 5.1 still priced at $10 per million input tokens and $50 per million output tokens. However, the cache read price has been cut by 75%, from $1 per million tokens to $0.25 per million tokens.
Performance improvements are clearly focused on long-term, multi-step Agent tasks, with Terminal-Bench-Science rising from 24.7% to 52.6% and AutomationBench from 17.1% to 31.4%.
What's perhaps more noteworthy than the model itself is the timing of its release – this is likely to be Anthropic's last major model release before going public.
In a few days, it will be September 7, the US Labor Day. According to plans previously disclosed by The Information, Anthropic, which has secretly submitted its IPO filing, may officially release its prospectus after the Labor Day and hold an investor day in mid-September, with the possibility of listing as early as late September to early October.
Meanwhile, the fallout from Anthropic's previous "raise in name only" quota adjustment also appeared in the comments section of the new model's release. On August 29, Anthropic announced that it would increase the "standard weekly quota" by 25% starting September 14, but this figure is calculated relative to the old standard prior to May; compared to the temporary quota currently in use by users, it actually decreases by 17%.
Shortly after Fable 5.1 went live, users pressed Anthropic on whether it could offer both more transparent and more generous session and weekly quotas.
Fable 5.1: Taking Agent to Longer Tasks
In terms of specifications, Fable 5.1 is not a major upgrade.
It continues Fable 5's 1 million token context window and 128K maximum output, with the base price remaining unchanged: $10 per million input tokens and $50 per million output tokens.
In the current flagship model lineup, Fable 5.1 is still significantly more expensive: based on 1 million inputs + 1 million outputs, Fable 5.1 is 2.5 times the current price of GPT-5.6 Sol and more than 10 times that of GLM-5.3.
Anthropic's official stance still recommends that most tasks start with Opus 5, which is half the price, and only switch to Fable 5.1 when faced with high-difficulty reasoning, long-duration Agent tasks, or when Opus 5 is insufficient even at high effort levels.
Notable changes have occurred in the cost of continuous running of the Agent.
The 5-minute and 1-hour cache write prices for Fable 5.1 remain at $12.5 and $20 per million tokens, respectively, but the cache read price has been reduced from $1 per million tokens in Fable 5 to $0.25, a 75% decrease.
Cache read can be understood as: after a segment of context has been processed by the model, subsequent requests do not need to re-read the complete input every time. For a ordinary Q&A session, this part may not be crucial; however, for an Agent that runs continuously for several hours, it will repeatedly read system prompts, code libraries, tool definitions, and previously processed context, and the longer the task and the more tools are called, the higher the cost of cache read will be.
Anthropic made calculations using actual usage data from a four-week period in August this year. In a token-based billing scenario, for the same task, switching from Fable 5 to Fable 5.1, the overall cost of typical workloads is expected to decrease by approximately 25%; for complex coding and highly agentized tasks with heavier context and more tool calls, the cost reduction can be as high as approximately 45%.
That is to say, although Anthropic did not make Fable 5.1 itself cheaper, it made it cheaper to "let it work for a long time".
However, being cheap is relative to the past, and the base price is still a factor. A developer commented in the release forum that even with a high cache hit rate, costs can still accumulate quickly, and a single complex coding task can cost between $20 and $50.
Additionally, this price reduction mainly occurs in API and other token-settled scenarios. For users who directly subscribe to Claude Code, another "quota account" system has just sparked controversy.
On August 29, the Claude Developers official account announced that starting from September 14, the "standard weekly quota" for Claude Code in Pro, Max, Team, and per-seat Enterprise packages will be permanently increased by 25%. However, previously Anthropic had temporarily raised this quota by 50%, and this temporary quota will remain in effect until September 14.
That is to say, if the original quota is considered 100, users are currently getting 150; after September 14, it will change to 125. Nominaly, this is a "permanent increase of 25%" compared to the old standard, but compared to the quota that users can actually use now, it is a decrease of about 17%. Anthropic later added a special explanation to confirm this calculation.
In the comments section following the release of Fable 5.1, users continued to ask Anthropic to provide "more transparent and generous" conversations and weekly quotas, and to stop playing similar verbal games.
However, judging by its performance shifts, Fable 5.1 increasingly looks like a model purpose-built for long-running tasks.
In the benchmarks released by Anthropic, the most significant improvement was seen in Terminal-Bench-Science 0.1, with Fable 5's score rising from 24.7% to 52.6%, more than doubling. The Terminal-Bench 4.0, which tests terminal environment agent capabilities, increased from 42.0% to 55.8%, while the AutomationBench, designed for complex business workflows, improved from 17.1% to 31.4%.
However, when switching to more traditional reasoning and Coding Benchmark, the improvement is not as dramatic.
Humanity's Last Exam saw an increase from 57.8% to 60.9% without tools, and from 63.8% to 65.0% with tools. CursorBench 3.2.0 rose from 70.5% to 73.4%, while the knowledge work benchmark GDPval-AA v2 increased from 1723 to 1853.
Overall, the improvements in Fable 5.1 are clearly focused on tasks that require the model to continuously invoke tools, process multiple steps, and check intermediate results before deciding on the next action, rather than uniformly distributing capabilities across all areas.
Early enterprise testing is mirroring this trend.
On its toughest browser Agent Benchmark, Browserbase found that Fable 5.1 completed 82% of tasks, Opus 5 completed 74%, and Fable 5 completed 57%.
In one of Ramp's tests, Fable 5.1 ran continuously for 38 hours without human intervention. It first discovered that the results of a previous machine learning experiment were actually affected by label bias, and then corrected the issue on its own, while launching six parallel experiments that ran all night, ultimately returning with new experimental results and suggestions for the next step. In another open-ended task, Ramp only asked it to find "problems that no one is responsible for, but are most worthy of solution," and it found an unhandled alert related to a production accident, retrieved logs, and provided a repair plan.
Canva's feedback was more focused on generation quality, stating that Fable 5.1's text expressions were easier to understand and more in line with writing standards. In a blind test comparing Fable 5.1 with Fable 5, Canva preferred the output of 5.1. Additionally, in Canva Code, Fable 5.1 directly created a rhythm game, not only generating levels and real music, but also synchronizing the gameplay rhythm with the musical beat. Canva said this was something that none of the other models they had tested could do.
From benchmarking to enterprise testing, the most notable improvement in Fable 5.1 is that it can now more stably complete complex tasks: it can sustain judgment, error correction, and progress over dozens of hours, and can also better organize text, code, music, and interactions in a single generation.
Claude enters the lab
Fable 5.1 and Mythos 5.1 appear to be two separate models, but Anthropic explicitly states in its official documentation:
Mythos 5.1 is "identical" to Fable 5.1, the difference being that it provides "more permissive safeguards" for vetted users.
In other words, the difference between the two lies in safety restrictions: Fable 5.1 is open to general users and enterprises, but is subject to additional restrictions when it comes to high-risk capabilities in cybersecurity and life sciences; Mythos 5.1, on the other hand, provides the same cutting-edge capabilities with fewer restrictions to verified cybersecurity and life sciences institutions through a trusted access program.
In the same benchmark, the two sometimes produce different results, mainly due to this reason. For example, on Terminal-Bench 4.0, which tests the ability to operate terminals and execute complex tasks for agents, Fable 5.1 scored 55.8%, while Mythos 5.1 reached 60.9%. According to Anthropic, this is not because Mythos has stronger underlying capabilities, but rather because some network security tasks triggered Fable's security restrictions; when these restrictions are reduced, the gap between the two should also narrow.
In this release, Anthropic devoted considerable attention to research capabilities, which were primarily showcased by Mythos 5.1.
One of the most eye-catching experiments was having Mythos 5.1 directly design protein binders.
Many modern medicines require combining with specific targets in the human body to complete blocking, activation, or delivery. Anthropic has connected Mythos 5.1 to open-source protein design and folding tools, allowing it to design binders on its own, and then sending the results to two external institutions for real-world experimental verification.
As a result, on 12 target sites, the binder hit rate of designs produced by Mythos 5.1 approached 50%. Anthropic said that the current hit rate in protein design is typically around 10% to 15%; on three of the target sites, the binders designed by the model had a binding affinity approximately 10 times higher than the best solution in Adaptyv Bio's protein design competition.
What's noteworthy here is not that Claude has learned to solve another biology problem, but that it has completed a more comprehensive scientific research process: using professional tools for molecular design and then sending the design to a real experimental environment for verification.
Something similar has happened in computer science.
Using radar data left by NASA's Magellan mission more than 30 years ago, Fable 5.1 trained a neural network to regenerate high-resolution elevation maps for about one-third of Venus's surface.
In the past, maps could distinguish terrain details at a scale of approximately 10 to 20 kilometers, while the new map narrows this down to 2 to 3 kilometers; Anthropic claims that the accuracy of height data has been improved by up to 25%. This map was subsequently made public under a Creative Commons license, with the hope of providing a reference for NASA's VERITAS and the European Space Agency's EnVision Venus exploration missions.
In addition, Mythos 5.1 has also done something even more engineering-oriented.
Biological research often requires repeated runs of deep learning models for proteins and genomes, with GPU speed directly determining how much time and money many experiments cost. Mythos 5.1 has customized GPU kernels for 7 open-source models and caches intermediate calculation results, ultimately increasing inference speed by up to 2.5 times while maintaining completely consistent output.
Anthropic estimates that these optimizations can reduce GPU costs by 30% to 60% in some whole-genome analysis tasks. Similar work typically requires a team of performance engineers to spend several weeks, while Mythos 5.1 achieved this in just a few days. Anthropic also plans to open-source these optimizations.
Taking these cases together, Anthropic's definition of Claude's research capabilities has undergone a change.
In the past, when discussing AI for Science, models were often used to read papers, retrieve materials, write code, or interpret results. With Fable 5.1 and Mythos 5.1, Claude has begun to engage in longer scientific research processes.
This road has even begun to transition from software to actual experimental equipment — just one week before the release of Fable 5.1, Anthropic opened a research preview of the Model Hardware Standard (MHS), with the goal of enabling AI agents to directly operate laboratory hardware such as microscopes, liquid handling devices, and robotic arms.
As capabilities advance towards scientific research and high-value tasks, Anthropic not only needs to face the security issue of "who to open it to", but is also increasingly concerned with another thing: how to prevent the core model capabilities from being "taken away" by others.
Fable 5.1 has added a stronger anti-distillation mechanism. In its official announcement, Anthropic directly named a practice where, in the past, API users could modify the historical context in multi-turn conversations while preserving Claude's previously generated thinking records. This allowed external models to collect Claude's reasoning trajectories in bulk and use them to train and distill their own models.
Starting from the release day of Fable 5.1, newly created API accounts can no longer operate in this way. Simply modifying the conversation context from before means that previously saved thinking transcripts can no longer be retained. Anthropic explicitly stated that this change is intended to block a "distillation technique that has been publicly documented"; existing accounts are temporarily unaffected, but when future models are released, this change will gradually be applied to all users.
The company is also bolstering its security system for enterprise clients. Anthropic's simultaneously launched Enterprise Frontier Safeguards (EFS) aims to resolve a real contradiction that has existed with Fable-level models: companies want to retain enough data to detect model misuse, but financial, medical, and legal enterprises are reluctant to have sensitive data stored by model companies for extended periods.
EFS' approach is to store the data required for monitoring directly in the customer's own controlled cloud infrastructure, rather than on Anthropic's servers. By default, human review is also carried out by the customer themselves. Anthropic said this mechanism was jointly developed with over 100 clients from industries including finance, healthcare, manufacturing, telecommunications, and law, and will be rolled out in phases starting from this autumn.
From protein design and Venus maps to vulnerability discovery, and then to reverse distillation and enterprise security, Fable 5.1 showcases more than just a model.
Anthropic is building a full suite of permissions, security, and infrastructure to bring cutting-edge capabilities into the worlds of research and enterprise, beyond its models.
After Fable 5.1, Anthropic is headed to Wall Street
If current plans remain unchanged, Anthropic may soon formally lay out its financials on Wall Street in just a few days.
The Information reported that Anthropic plans to publicly release its IPO prospectus after the US Labor Day on September 7 and hold an investor day in mid-September, with the possibility of listing as early as late September to early October. The company had secretly submitted its IPO filing in June.
Coming at this time, Fable 5.1 naturally serves as a demonstration of Anthropic's capabilities ahead of its public listing.
After all, once the prospectus is officially made public, what Wall Street needs to determine is not whether Claude can achieve high scores in a few more Benchmarks, but how much Anthropic is actually worth.
In May, Anthropic's latest private funding round valued the company at $965 billion. By August, several investors interviewed by the Financial Times already expected the company's valuation at IPO to exceed $2 trillion. Investors project annualized revenue could reach $100 billion to $120 billion by year-end, with one investor even calculating a $3 trillion figure based on the company's growth trajectory and a 30x revenue multiple. Of course, these are currently just investors' own valuation models—the final offering price has yet to be determined, and plans could change at any time.
" top of a, top of a top of a top of a a top of a top of. " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " that " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " a top of " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " "." " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " a " " " " " " " " " " " " " " " " " " " " " " " " " " " a " " " " " " " "
According to internal financial forecasts obtained by Reuters in August, Anthropic expects its revenue to reach $190 billion to $200 billion by 2028. The company's annualized revenue run rate has also grown from approximately $9 billion at the end of 2025 to $47 billion in May this year, and further exceeded $65 billion by the end of July. Documents obtained by Bloomberg also show that the company's preliminary revenue in the second quarter exceeded $11.5 billion, more than 14 times that of the same period last year, and it recorded positive adjusted operating profit for the first time.
While Wall Street has already begun to price it based on its value over the next few years. Reuters said that investment banks and investors are using 2028 revenue forecasts to calculate Anthropic's enterprise value-to-revenue multiple, with Palantir, Cloudflare, and SpaceX serving as references in different dimensions.
As it approaches its listing, Anthropic is still spending heavily on buying GPUs, training models, conducting inference, and expanding its team.
One week before the release of Fable 5.1, Anthropic was revealed to have signed a six-year, $45 billion computing power contract with Nscale, renting approximately 460MW of data center capacity in West Virginia. A few days later, Reuters disclosed that Anthropic had signed a cloud computing contract worth around $35 billion with Lambda, which is backed by Nvidia, corresponding to approximately 350MW of data center capacity in Nueces County, Texas.
These two recently exposed computing power commitments alone have reached $800 billion (approximately 5,377 billion yuan).
This also gives another layer of meaning to the seemingly scattered changes that came before Fable 5.1.
This a top of a top of a top of a " top of top of top of a top of top of a top of the a. " ( that a top of a also a top of a top of a. " ( in a top of a top of of a. " " ( in a top of a top of a. " " " " " " " " " " " " " " " " " " " " of a. " " " " " " " " " " " " " " " " " " " " of a. " " " " " " " " " " " " " " " " " " " " " " " " " " " " a no of a. " " " " " " " " " " " " " in a. " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " ". " " " " " " " " " " " " " " in a. of a. " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " "
On the other hand, merely proving that a "model is smarter" is obviously no longer enough.
Protein design, scientific computing, enterprise Agent, network security, as well as Enterprise Frontier Safeguards mentioned in the second part, are all driving Claude in one direction: towards problems that enterprises are originally willing to spend a lot of money to solve.
Fable 5.1's cache read price cut follows the same logic. Rather than competing on low base-token pricing, Anthropic targeted long-running agents, cutting real-world costs for typical workloads by roughly 25% and by as much as 45% for highly agentic tasks.
The model is stronger, and Agent can work for longer periods; the longer the working time, the more it is necessary to reduce the computing power cost per unit task.
This may be the most crucial math problem Anthropic needs to solve after going public. Growth has been rapid enough, but the question is whether revenue can ultimately outpace costs, which are also ballooning at a rate of hundreds of billions of dollars.
The value of Fable 5.1 is not just about Anthropic achieving another impressive Benchmark score before its public listing. From longer-term agents, to scientific research and high-value enterprise tasks, to cost reduction, security, permissions, and infrastructure, Anthropic is attempting to demonstrate a more complete chain.
Model capabilities can be turned into jobs, jobs can be turned into income, and income can ultimately outpace computing power.
In a few days, after Anthropic releases its prospectus, this narrative will no longer be just a product story for a model company.
It will also become the basis for Wall Street to price Anthropic.