The AI race seems to be changing its course. It is no longer only about which company has the smartest model. Now, AI companies are trying hard to figure out how cheaply they can deliver powerful AI models. At a time when Chinese AI companies are offering their AI capabilities at lower prices. That puts pressure on companies such as Anthropic, OpenAI and Google to find ways to make powerful AI more affordable. And Anthropic’s latest move is Claude Fable 5.1.
Anthropic cuts AI costs
Anthropic has introduced Claude Fable 5.1 and Claude Mythos 5.1, which use the same underlying model but come with different safety controls. Fable 5.1 is generally available, while Mythos 5.1 is restricted to trusted programmes focused on areas such as cybersecurity and life sciences. Now the notable part here is the pricing.
Anthropic says Fable 5.1 will cost around 25 per cent less than Fable 5 for typical workloads. For highly agentic workloads, where AI works through long and complicated tasks, the savings can reach around 45 per cent.
However, Anthropic has not reduced the model’s headline API price. Fable 5.1 still costs $10 per million input tokens and $50 per million output tokens.
Instead, the company has cut the price of cached input. This is information the model has already processed and can reuse later. The cache-read price has fallen from $1 to just $0.25 per million tokens, a 75 per cent reduction.
That matters because AI agents often revisit the same code, documents, instructions and conversation history. Anthropic says the cheaper caching can reduce costs by around 25 per cent for typical workloads and up to roughly 45 per cent for highly agentic work.
For your knowledge, DeepSeek’s V4 Flash, for instance, is priced at $0.44 per million input tokens and $1.32 per million output tokens during peak hours, while its off-peak rates are even lower. And MiniMax’s M2.7 is also priced at $0.30 per million input tokens and $1.20 per million output tokens.
The real AI race is now about cost per task
This shift is becoming important as AI moves beyond chatbots. A chatbot may answer a question in seconds. An AI agent can spend hours reading documents, writing code, testing its work and trying again when something goes wrong. The more work it does, the more computing power it needs.
That means companies increasingly have to ask not just how much a model costs per million tokens, but how much it costs to complete a task.
Anthropic is also positioning Fable 5.1 as a model built for this kind of long-running work. The company says it scored 52.6 per cent on Terminal-Bench-Science 0.1, compared with 24.7 per cent for Fable 5 and 29 per cent for Opus 5. On AutomationBench, it scored 31.4 per cent, compared with 17.1 per cent for Fable 5.
Early customers have reported similar improvements. Investment firm Millennium said Fable 5.1 found the cause of a rare software crash that engineers had failed to explain for four to five years. Ramp also said it ran the model unattended for 38 hours, allowing it to run multiple experiments before returning with results.
But greater autonomy also brings bigger risks. Anthropic recently disclosed incidents involving earlier Claude models that, during permissive cybersecurity testing, reached real-world systems. One model accessed production data, while another uploaded malicious code to a real software repository.
Anthropic says these incidents happened during testing with safeguards disabled. Still, they highlight a problem that the entire AI industry will have to deal with as models become more autonomous: the more an AI can do on its own, the more important its permissions and safety controls become.
Anthropic says Fable 5.1 comes with improved safeguards that produce around 60 per cent fewer interventions per Claude Code session.
The company is also introducing Enterprise Frontier Safeguards, which allows businesses to keep monitoring data inside their own cloud infrastructure while still allowing Anthropic’s systems to detect potential misuse.
The bigger takeaway is simple. AI companies are no longer competing only to build smarter models. They are competing to make those models cheaper, faster and practical enough to run at scale.
And with cheaper Chinese models adding pressure, that price war is only getting started.
– Ends
Leave a Reply