Anthropic rolled out Claude Fable 5 on Tuesday, its newest general-use AI model and the company’s most capable publicly available system to date. The Mythos-class model beats rivals on coding, finance, and vision benchmarks, according to internal tests. Pricing is set at less than half the cost of Anthropic’s own Claude Mythos Preview, a move that could shake up the crowded large-language-model market.
What Claude Fable 5 brings
Fable 5 is built on the Mythos architecture, the same family as the earlier Mythos Preview. But Anthropic says this version pushes performance further across several key domains. In coding tasks it outperforms comparable models from OpenAI, Google, and Meta, the company reported. Financial analysis benchmarks also show a lead, and vision-based tasks — such as interpreting charts and diagrams — score higher than previous Anthropic releases and rival systems.
The model is available starting Tuesday through Anthropic’s API and its consumer chatbot, Claude. Users can access it immediately, though the company hasn’t disclosed a specific usage cap or rate limit for the launch.
Price cut changes the math
The most striking detail in Tuesday’s announcement is the price. Claude Fable 5 costs less than half what Claude Mythos Preview charges per token. Anthropic didn’t release exact per-token figures in the launch materials, but the discount is significant enough that the company framed it as a strategic advantage. For developers and businesses that run large volumes of queries, the savings could be substantial.
The lower price also puts pressure on competitors. OpenAI’s GPT-4o and Google’s Gemini 1.5 Pro carry similar per-token costs to the older Mythos Preview, meaning Fable 5 undercuts them on price while claiming superior performance on certain benchmarks. Whether real-world workloads match the internal tests remains to be seen, but the pricing alone is likely to draw attention from enterprise buyers.
Benchmark claims vs. real competition
Anthropic provided comparisons on three benchmark categories: coding (using HumanEval, SWE-bench, and internal evaluations), finance (a mix of financial reasoning and summarization tasks), and vision (image captioning and visual QA). In each, Fable 5 scored higher than GPT-4o, Gemini 1.5 Pro, and Meta’s Llama 3.1 405B, according to Anthropic’s data.
Independent verification hasn’t been published yet. Benchmark results from the company that builds the model are standard in the industry, but third-party audits — like those run by Stanford’s HELM or the LMSYS Chatbot Arena — often carry more weight with skeptical developers. Anthropic didn’t say when or if it would submit Fable 5 to those evaluations.
The model also supports multimodal inputs, meaning users can upload images alongside text. That feature wasn’t available in the Mythos Preview release, which was text-only. Anthropic says vision performance on Fable 5 exceeds that of dedicated vision models from other labs on the tasks tested.
What comes next
Anthropic hasn’t announced a timeline for replacing the older Mythos Preview tier. For now both models remain available, but the steep price gap and performance gains make Fable 5 the obvious choice for new projects. Existing users on Mythos Preview may want to test Fable 5 and compare results before migrating production workloads.
The launch lands as the AI arms race enters a new phase — not just about raw capability, but about affordability at scale. Anthropic’s bet is that a cheaper, stronger model will pull in customers who balked at previous pricing. Whether that bet pays off depends on how Fable 5 performs outside the lab, in the messy, unpredictable world of real applications.
