Anthropic shipped Claude Fable 5 on June 9, 2026 — the first publicly available Mythos-class model. If you use Claude in any production workflow, there are three things to understand immediately: what it costs, what the June 22 deadline means for your team, and whether it actually belongs in your stack.
What Fable 5 is
Fable 5 is the Mythos model made available to the public, with safety classifiers wrapped around the capabilities that Anthropic was holding back. The unrestricted version — Claude Mythos 5 — stays locked inside Project Glasswing and is available only to approved partners in cybersecurity and critical infrastructure.
The two models share the same underlying weights. Pricing is identical: $10 per million input tokens, $50 per million output tokens. That is exactly double Claude Opus 4.8’s rate of $5/$25, and less than half of Mythos Preview’s original $25/$125 pricing.
On benchmarks, Fable 5 posts 80.3% on SWE-Bench Pro — the highest score any model has achieved on that coding benchmark. For operators running AI coding workflows, software automation, or agentic document-processing pipelines, that number matters more than general chat quality scores.
The June 22 deadline
Here is the detail that is easy to miss: Fable 5 is included in Claude Pro ($20/month), Max 5x ($100/month), Max 20x ($200/month), Team Standard, Team Premium, and seat-based Enterprise plans only through June 22, 2026. Starting June 23, access requires usage credits — the consumption billing that kicks in once your included plan limits are exhausted.
Anthropic is framing this as a capacity constraint, not a permanent pricing change. The company says it intends to restore Fable 5 as a standard subscription feature once more compute is online. But “we’ll try to restore it” is not a guarantee, and teams that have built workflows assuming Fable 5 is within their monthly plan cost need to account for the credit overhead starting next week.
A Fable 5 session counts as approximately 2x the usage weight of an Opus 4.8 session against any plan limits. Run the math on your current usage and figure out what the credit tab looks like.
API pricing mechanics
For teams on direct API billing, there is no cliff: Fable 5 is available at $10/$50 today with no subscription deadline. Three pricing notes worth tracking:
Prompt caching saves up to 90% on cached input. If your workflow sends repetitive system prompts or long documents, turn on caching before migrating to Fable 5. This single optimization can bring effective input cost below Opus 4.8’s uncached rate.
Batch API pricing is 50% off standard — $5/$25 per million tokens. For non-realtime workloads (document processing, bulk classification, analysis pipelines), batch mode cuts your cost in half.
US-only inference is available at a 1.1x multiplier. For latency-sensitive applications, the US routing option trades a 10% cost premium for lower latency and data residency certainty.
Fable 5 is available via Claude API directly, AWS Bedrock, Google Vertex AI, and Microsoft Foundry. If your infrastructure is already on one of those platforms, there is no migration step — just update the model ID to claude-fable-5.
Whether it belongs in your stack
The pricing position — double Opus 4.8 — means Fable 5 should not replace Opus across the board. It belongs in high-value inference tasks: complex reasoning, multi-step code generation, document-level analysis where quality directly affects downstream decision quality.
For high-volume tasks where Opus 4.8 is already accurate enough — sentiment classification, entity extraction, summarization of structured content — there is no ROI case for switching. The 2x cost premium only pays off when Fable 5’s capability delta produces a measurable quality improvement in that specific task.
What to do this week
Before June 22: Audit which of your Claude workflows are on subscription plans and estimate the credit cost if Fable 5 stays behind usage credits. Calculate the breakeven point where upgrading to API billing outperforms credit purchases.
Before end of June: Run Fable 5 side-by-side against Opus 4.8 on your specific tasks. If you have coding or agentic workflows, the SWE-Bench improvement is real — 80.3% is a meaningful gap from the previous generation. If you have classification or extraction workflows, the upgrade may not close.
Keep an eye on the Anthropic changelog for any update to the June 22 date. Anthropic has hinted it may extend the free window if capacity allows, but plan for the deadline as stated.