What happens to Claude Fable 5 on June 22, 2026?

Claude Fable 5 free access on Anthropic's Pro, Max, Team, and seat-based Enterprise plans ends June 22, 2026. From June 23, Fable 5 usage on those plans requires credits, with each Fable 5 query consuming roughly 2x the credit allowance of an Opus 4.8 query. Anthropic has said it plans to restore subscription access once compute capacity allows, but no specific date has been announced. For most production teams using Fable 5 heavily, the credit-based pricing materially changes the cost equation. The straightforward implication: you need a routing plan in place before June 23.

What is the best routing plan for the June 23 cutover?

Three-tier routing. Tier 1 (5-15% of work, hardest tasks): Claude Fable 5 via credits or Anthropic API. Use only when quality matters more than cost — hardest SWE-Bench Pro-class problems, vision-required tasks, long-horizon agentic loops where a model error compounds. Tier 2 (50-70% of bulk agentic coding): GLM-5.2 via OpenRouter at $1.40/$4.40 per M tokens, or self-hosted. Best capability per dollar among open-weight options as of June 19, 2026. Tier 3 (15-25% of routine work): a small coding model — North Mini Code, Qwen 3.6 35B-A3B, or Devstral Small 2. Use Claude Opus 4.8 as the fallback inside the Anthropic ecosystem when Fable 5 credits run out.

Should I switch entirely off Anthropic on June 23?

For most teams, no. Anthropic's product surface — Claude Code, Claude Desktop, MCP ecosystem, sub-agents, dynamic workflows — is the most mature agentic coding stack available. The cost-effective pattern is to keep Claude Code as the IDE and harness, but route inside Claude Code to non-Anthropic models for bulk work. Cursor and Claude Code both support custom model routing via OpenAI-compatible endpoints. You can keep the Anthropic workflow, swap the model for GLM-5.2 or Kimi K2.7 Code, and pay 4-7x less for the bulk of work while keeping Fable 5 / Opus 4.8 for the hardest 10-20%.

What if GPT-5.6 ships during the cutover window?

Polymarket assigns 83% probability to GPT-5.6 launching between June 22-28, 2026. If GPT-5.6 ships with materially better coding capability at competitive pricing, it joins Tier 1 (or possibly takes the bulk-agentic Tier 2 slot if cheap enough). The right approach: keep your routing infrastructure model-agnostic (use OpenRouter or a custom router that abstracts the underlying model), so you can swap GPT-5.6 in without rewriting your stack. If GPT-5.6 launches at GPT-5.5 pricing levels or below with Fable-5-class coding, it likely becomes the new bulk default. Watch the launch window carefully.

Quick Answer

Claude Fable 5 Paywall June 22: Routing Plan for the June 23 Cutover

Published: June 19, 2026

Claude Fable 5 Paywall June 22: Routing Plan for the June 23 Cutover

On June 22, 2026, free Claude Fable 5 access on Anthropic’s Pro, Max, Team, and Enterprise plans ends. From June 23, Fable 5 queries require credits — and each Fable 5 query consumes roughly 2x the credit allowance of an Opus 4.8 query. For teams running Fable 5 heavily, the bill on June 23 will be a surprise unless you plan now. Here is the concrete routing playbook.

Last verified: June 19, 2026. You have 3 days.

TL;DR

June 22, 2026: Last day of free Fable 5 on Pro/Max/Team/Enterprise plans.
June 23, 2026: Credit-based pricing on Fable 5 begins (~2x Opus credits per query).
The fix: Three-tier routing — Fable 5 for hardest 5-15%, GLM-5.2 for bulk 50-70%, small model for routine 15-25%.
Keep your IDE/harness. Cursor and Claude Code both support custom model routing.
Watch for GPT-5.6. 83% Polymarket probability of June 22-28 launch may reshape the routing math within days.

Step 1: Audit what you’re actually using Fable 5 for

Most production stacks use Fable 5 for far more than they need to. Categorize your last 7 days of queries into three buckets:

Bucket	Typical share	Examples
Hardest tasks (Tier 1)	5-15%	Hard SWE-Bench Pro-class problems, vision-required code reasoning, long agentic loops where quality compounds, novel architectural decisions
Bulk agentic (Tier 2)	50-70%	Routine repo edits, function implementation, refactors, test generation, multi-step but bounded agentic flows
Routine (Tier 3)	15-25%	Boilerplate, single-function edits, structured-output generation, formatting fixes, edge cases

Only Tier 1 needs Fable 5’s quality ceiling. Tier 2 and Tier 3 are where the 4-10x cost savings come from routing elsewhere.

Step 2: Pick the Tier 2 model

Option	Cost per M (in/out)	Strengths	Weaknesses
GLM-5.2 (OpenRouter)	~$1.40 / $4.40	#1 open-weight on AA Intelligence Index (51), 1M context, MIT license	No vision, token-hungry (~43k output tokens per task)
Kimi K2.7 Code (Moonshot)	$0.95 / $4.00	Strongest MCP tool-use scores (Atlas 76.0, Mark Verified 81.1), vision input	Slightly lower raw intelligence than GLM-5.2
DeepSeek V4 Pro	<$1.00 / <$2.00	Cheapest, biggest active parameter count, most mature ecosystem	Behind GLM-5.2 on Intelligence Index
Claude Opus 4.8 (credits)	Internal credit cost	Same vendor, same harness, mature workflow	Still expensive relative to open weights

Recommendation as of June 19, 2026: GLM-5.2 via OpenRouter is the strongest default for most teams. Kimi K2.7 Code is the better pick if your stack is MCP-tool-heavy or needs vision. DeepSeek V4 Pro is the right pick if cost is the dominant constraint.

Step 3: Pick the Tier 3 model

Option	Cost per M (in/out)	Strengths
North Mini Code (Cohere)	varies	Apache 2.0, agent-harness-robust, Western sovereign, 2.8x Devstral throughput
Qwen 3.6 35B-A3B	varies	Highest small-coding-model AA Coding Index (35.2)
Devstral Small 2	varies	OpenCode incumbent, Apache 2.0

For most teams, North Mini Code or Qwen 3.6 35B-A3B are roughly interchangeable on capability. Pick on procurement and ecosystem fit.

Step 4: Wire up the routing inside Claude Code or Cursor

Both Claude Code and Cursor support custom model routing through OpenAI-compatible endpoints. The minimum config:

OpenRouter API key. Sign up at openrouter.ai. Add credit. Get your API key.
Configure Claude Code or Cursor to add a custom OpenAI-compatible endpoint at https://openrouter.ai/api/v1 with your OpenRouter key.
Add model routes for z-ai/glm-5.2, moonshotai/kimi-k2-7-code, deepseek-ai/deepseek-v4-pro, and your Tier 3 pick.
Define a routing rule (manual or automatic) inside your harness — for example, default to GLM-5.2 for new tasks; only escalate to Fable 5 when you explicitly mark a task as hard.

For Cursor: use the model picker UI to add custom OpenAI-compatible models.

For Claude Code: configure custom endpoints in your .claude/settings.json or via claude config set.

For terminal/MCP setups: use litellm or a similar router to abstract endpoints.

Step 5: Set a Fable 5 credit budget

Even with routing, you will still use Fable 5 for the hardest 5-15%. Set a monthly credit budget that matches your Tier 1 expected usage. Two patterns:

Conservative: $50-100/month of Fable 5 credits for a single dev doing heavy coding. Covers the hardest tasks; bulk goes to GLM-5.2.
Aggressive: $300-500/month for a senior dev or small team. Covers more Tier 1 work without manual escalation gates.

Anthropic has said they plan to restore subscription access “once capacity allows” — but no date. Plan as if credits are the new normal.

Step 6: Watch for GPT-5.6 (June 22-28 window)

Polymarket assigns 83% probability to GPT-5.6 launching between June 22-28, 2026. If it ships:

At Fable 5 quality + Fable 5 pricing: No change to your routing — Fable 5 stays in Tier 1.
At Fable 5 quality + GPT-5.5 pricing or lower: GPT-5.6 likely takes the bulk Tier 2 slot from GLM-5.2.
At sub-Fable 5 quality + much lower pricing: GPT-5.6 joins Tier 2/3 mix; no significant routing change.

Keep your routing abstraction model-agnostic so a swap is one config change, not a rewrite.

The “do nothing” alternative

If you ignore the June 23 cutover and keep using Fable 5 for everything, the most likely outcomes:

Your monthly Anthropic bill increases 3-10x relative to the free-Fable-5 period.
Or your credits run out mid-month and you fall back to Claude Opus 4.8 automatically (Anthropic’s fallback).
Or you start hitting rate limits if Anthropic tightens credit-based capacity.

None of these break your workflow. They are just more expensive than the routing pattern above.

The honest read

The June 23 transition is not a crisis. It is a forcing function. Most teams using Fable 5 heavily in the June 9-22 free window were overusing it relative to what their workloads actually need. The routing plan above is what most teams should have been running anyway; the paywall just makes the cost incentive obvious.

Set up the routing this week. By June 23, you have a stack that delivers Fable 5 quality on hardest tasks and 4-10x cost savings on the bulk. That is the new normal until subscription access returns — or until GPT-5.6 changes the equation again, possibly within days.