# Anthropic Cuts Prices 20 Percent With Claude Fable 5.2 and Reports 49.1 Percent on FrontierCode 2.0

Fable 5.2 lists at $8 per million input tokens and $40 per million output, and Anthropic's headline coding score is measured at maximum effort and drops to 37.6 percent at medium.

- Published: 2026-11-14T06:20:14.641Z
- Canonical: https://polylog.news/ai/2026-11-14/anthropic-cuts-prices-20-percent-with-claude-fable-5-2-and-r
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Anthropic News](https://www.anthropic.com/claude-fable-and-mythos-5-2), [DataCamp](https://www.datacamp.com/blog/claude-fable-5-2-and-mythos-5-2), [Implicator](https://www.implicator.ai/anthropic-fable-5-2-mythos-5-2-price-cut/), [VentureBeat](https://venturebeat.com/technology/anthropic-releases-claude-fable-5-2-and-mythos-5-2-pairing-new-record-high-benchmarks-with-20-api-cost-reduction)

Anthropic [released Claude Fable 5.2 and Claude Mythos 5.2](https://www.anthropic.com/claude-fable-and-mythos-5-2) on November 9 and described them as its most capable models for coding, finance and the sciences. Mythos 5.2 is the same underlying model with fewer safeguards. Access to it is restricted to vetted organizations.

The vendor-reported coding gains are large. According to the table compiled by [DataCamp](https://www.datacamp.com/blog/claude-fable-5-2-and-mythos-5-2), Fable 5.2 scores 49.1 percent on FrontierCode 2.0, up from 32.4 percent for Fable 5.1 and above Opus 5.5 at 45.5 percent. It scores 69.9 percent on FrontierSWE v2 against 62.3 percent for Opus 5.5. Anthropic also reports 41.3 percent on Remote Labor Index, against 21.3 percent for Opus 5.5, and 89.3 percent on OSWorld 2.1. On the restricted model, CVE Rediscovery rises to 89.6 percent from 78.3 percent for Mythos 5.1.

Several of these figures hold up less well once their conditions are specified. The FrontierCode figure is reported at the highest effort setting. Third-party coverage gives 37.6 percent at medium and 45.3 percent at high, so cost per solved task depends on the effort setting. On Terminal-Bench 4.0, Fable 5.2's 67.6 percent against Opus 5.5's 66.4 percent falls within margins of error. One independent HealthBench run put Fable 5.2 at 58.8 percent, below Opus 5.5 at 64.3 percent. Most published figures trace back to Anthropic's own launch materials and have not been independently reproduced.

Pricing moved in the opposite direction from the capability claims. Fable 5.2 [costs $8 input and $40 output per million tokens, down from $10 and $50](https://www.implicator.ai/anthropic-fable-5-2-mythos-5-2-price-cut/), and cache and batch rates fell by the same 20 percent. Mythos 5.2 carries the same list rates. Implicator notes that Anthropic's claim of output up to 30 percent faster than 5.1 could not be verified at publication.

## What this means

Anthropic is cutting price on its flagship while it prepares a public listing, which shows the company judges market share in coding agents to be worth more than margin per token. Its S-1 shows compute as nearly all of its cost of revenue, so the cut is either a bet on serving efficiency or a deliberate margin sacrifice. Rival coding models at OpenAI and Google, and open-weight coding models priced far lower, face pressure to respond. Engineers who route traffic by cost per solved task gain most, but only if they measure at the effort level they actually run, since the headline score sits at the most expensive setting. The consensus reads the release as a capability jump. The more durable change is the price, because the benchmark lead is narrow on Terminal-Bench and reversed on one HealthBench run. The likelier outcome is a matching price move from at least one other frontier vendor within weeks. That depends on Anthropic's 30 percent speed claim holding up, which would support the lower price. It would be proven wrong if independent agentic-coding runs at medium effort show Fable 5.2 no better than Opus 5.5.

## What to watch

- Independent FrontierCode and FrontierSWE runs at medium and high effort, which would show whether the gain survives at settings teams can afford.
- Whether OpenAI or Google reprice their coding-tier models in the next few weeks, which would confirm price as the competitive axis.
- Which organizations Anthropic admits to Mythos 5.2, since a gated offensive-security model creates a separate market with its own buyers.