Is Claude Fable 5 now the default in Claude Code?

Yes for Pro and Max subscribers as of June 9, 2026. When you launch Claude Code with a claude.ai Pro or Max account linked, Fable 5 is now selected by default, with automatic fallback to Opus 4.8 when your credit balance is running low. Free-tier and pay-per-API users keep their previous defaults — you can opt in to Fable 5 with --model claude-fable-5 or by editing ~/.claude/config.json.

What is the June 22, 2026 Claude credits change?

On June 22, 2026, Anthropic activates a usage-credits system for claude.ai Pro, Max, and Team subscriptions, replacing the previous message-count quotas. Each API call now consumes credits proportional to actual token cost — Fable 5 consumes credits roughly 2x faster than Opus 4.8. Effective monthly quota for heavy Claude Code Pro users drops meaningfully if they stay on Fable 5. Power users may need to upgrade from Pro ($20) to Max ($100) to maintain previous throughput.

How do I keep Claude Code cost predictable after June 22?

Three tactics. First, set fallback_models in ~/.claude/config.json so Claude Code automatically falls back to Opus 4.8 when Fable 5 quota is tight. Second, use the orchestrator-subagent pattern: Opus 4.8 orchestrator + Haiku 4.5 subagents via Dynamic Workflows is ~15x cheaper than monolithic Fable 5 for similar quality. Third, pin specific workflows to specific models — Fable 5 for hard agentic coding, Opus 4.8 for routine work, Haiku 4.5 for high-volume edits. Monitor with 'claude usage' to see credit burn rate.

Should I upgrade from Claude Pro to Claude Max for Fable 5?

Worth considering if you run Claude Code 4+ hours per day. Pro at $20/mo gives a baseline quota that Fable 5 burns through quickly. Max at $100/mo gives 5x the credits with priority access. The math: if you're hitting Pro quota limits weekly on Opus 4.8, you'll hit them daily on Fable 5. If your workload is mostly routine coding, stay on Pro and use Opus 4.8 as the default. If you do long agentic-coding runs daily, Max is worth the upgrade. Or skip subscriptions entirely and use a pure API key with pay-as-you-go billing.

Quick Answer

Claude Fable 5 Rollout: Claude Code Pro/Max Credits June 22

Published: June 10, 2026

Claude Fable 5 Rollout: Claude Code Pro/Max Credits June 22

Anthropic flipped two big switches in three weeks. June 9, 2026: Claude Fable 5 became the default model in Claude Code for Pro/Max subscribers. June 22, 2026: the new usage-credits system kicks in and Fable 5 burns 2x the credits of Opus 4.8. Here’s how to manage the combination.

Last verified: June 10, 2026

The two changes

Date	Change	Effect
June 9, 2026	Claude Fable 5 launches publicly; becomes Claude Code default for Pro/Max	Better quality, 2x API cost
June 22, 2026	New usage-credits system replaces message-count quotas for claude.ai Pro/Max/Team	Heavy users hit limits faster

The combination is what matters: if you migrated to Fable 5 as the default after June 9 and you have a claude.ai Pro/Max subscription, the June 22 credits change effectively halves your monthly throughput compared to staying on Opus 4.8.

What Claude Fable 5 brings

Best-in-class agentic coding — SWE-Bench Pro 80.3% (vs Opus 4.8 71.4%, GPT-5.5 58.6%)
Frontier reasoning — GPQA Diamond 87.8%, AIME 2025 96.2%
1M-token context (vs Opus 4.8’s 500k)
128k token max output per request
Same MCP tool composition as Opus 4.8 — no migration cost

Coverage: What is Claude Fable 5?

What it costs

Model	Input ($/MTok)	Output ($/MTok)	Credit multiplier
Claude Fable 5	$10	$50	~2.0x
Claude Opus 4.8	$5	$25	1.0x (baseline)
Claude Haiku 4.5	$0.80	$4	~0.16x

Fable 5 is exactly 2x the API cost of Opus 4.8. The usage-credits multiplier roughly tracks that — so a workload that used to fit in your Pro quota on Opus 4.8 will use about 2x the credits on Fable 5.

The June 22 credits change explained

Before June 22, claude.ai Pro/Max/Team had a fuzzy “message limit” — capped roughly by token budget but not exposed. Heavy users often hit invisible throttles unpredictably.

After June 22, you’ll see explicit credit balance:

$ claude usage
Plan: Claude Pro ($20/mo)
Credits this billing cycle:
  Used:      8,420 / 12,000
  Reset:     June 30, 2026
  Burn rate: ~340 credits/day (model: claude-fable-5)
  Estimated runway: 10.5 days

Each model has a credit cost roughly proportional to API token cost:

Fable 5 / Mythos 5: ~2.0 credits per 1k input + 10 credits per 1k output
Opus 4.8: ~1.0 credits per 1k input + 5 credits per 1k output
Haiku 4.5: ~0.16 credits per 1k input + 0.8 credits per 1k output

(Exact ratios may vary by tier and time of day. Anthropic has not yet published a final published-rate table — these are working approximations.)

Practical impact on Claude Code users

Light user (1–2 hours/day of Claude Code)

Pro ($20/mo) + Opus 4.8 default: comfortable
Pro ($20/mo) + Fable 5 default: might hit quota in last week of month
Recommendation: keep Opus 4.8 default; switch to Fable 5 only for hard tasks

Heavy user (4+ hours/day, big agentic-coding workflows)

Pro ($20/mo) + Fable 5 default: will hit quota in week 2 or 3
Max ($100/mo) + Fable 5 default: comfortable but not unlimited
Recommendation: upgrade to Max, use orchestrator-subagent pattern to amplify quota

Pure API user (no claude.ai subscription)

Pay-as-you-go API: Fable 5 doubles your bill on migrated workloads
Cost-conscious option: use Opus 4.8 orchestrator + Haiku 4.5 subagents (15x cheaper than monolithic Fable 5)
Recommendation: be deliberate about which workloads merit Fable 5

How to manage cost — concrete tactics

1. Set explicit defaults and fallbacks

Edit ~/.claude/config.json:

{
  "model": "claude-fable-5",
  "fallback_models": ["claude-opus-4-8", "claude-haiku-4-5"],
  "auto_fallback_on_quota": true,
  "max_credits_per_session": 500
}

2. Pin models per workflow

Use Claude Code’s --model flag for the right model per task:

# Hard refactor — use Fable 5
claude --model claude-fable-5 "Refactor src/api/ to use new error types"

# Routine edit — use Opus 4.8
claude --model claude-opus-4-8 "Fix the lint errors in src/utils/"

# Bulk classification — use Haiku 4.5
claude --model claude-haiku-4-5 "Classify these 100 issues by severity"

3. Use the orchestrator-subagent pattern

For agent workloads, this is the biggest cost multiplier:

# Orchestrator: Opus 4.8
# Subagents: Haiku 4.5 (cheap fan-out)
# Total cost: ~5–10% of monolithic Fable 5
claude --orchestrator claude-opus-4-8 \
       --subagent claude-haiku-4-5 \
       "Audit the entire src/ for dead code"

Coverage: Dynamic Workflows 1000-subagents cap explained

4. Monitor burn rate

claude usage --window 24h
claude usage --by-model
claude usage --by-skill

If burn rate exceeds your budget, switch defaults or upgrade tier.

5. Cache aggressively

Fable 5 supports prompt caching at the standard 90% read discount. For repeated runs (codebase context, prompt scaffolding), this can cut effective cost by 50–80%.

claude --cache-system "Always read this file first" \
       --cache-tools "Use these MCP tools" \
       "Now do the task"

When to upgrade Pro → Max

Indicator	Action
Pro quota exhausted by day 20 of billing cycle	Consider Max
Pro quota exhausted by day 10 of billing cycle	Upgrade to Max
Pro quota exhausted by day 25	Stay on Pro
Fable 5 default + Pro + Heavy agent runs	Max is likely cheaper than pay-as-you-go API
Pure API user with $150+/mo bill	Try Max as a base + minimal API overflow

When to downgrade Fable 5 → Opus 4.8

Practical cases where staying on Opus 4.8 is the smart play:

Routine code edits, refactors, lint fixes — quality gap doesn’t justify 2x cost
High-volume orchestrator workloads — fan-out economics favor Opus + Haiku
Latency-sensitive interactive coding — Opus is snappier
Cybersecurity / red team work — Fable 5 refuses more, Opus 4.8 handles defensively
Cost-constrained teams — Opus 4.8 is the budget-friendly frontier choice

Coverage: Claude Fable 5 vs Opus 4.8: Should you upgrade?

What’s likely next

Likely change	Timing
Anthropic publishes formal credit-rate table	By June 22, 2026
Opus 5.0 launch	Q3 2026 (Fable-5-level quality at half the price)
Haiku 5 launch	Q3 2026 (cheap fan-out gets cheaper)
Mythos 5 trusted-access program for verified researchers	Q3 2026
Refinement of Fable 5 false-positive refusals	2–4 weeks
Cost-tier auto-routing in Claude Code	Q4 2026

Recommended setup for most developers as of June 10, 2026

{
  "model": "claude-opus-4-8",
  "fallback_models": ["claude-haiku-4-5"],
  "model_overrides": {
    "hard-refactor": "claude-fable-5",
    "research-grade-reasoning": "claude-fable-5",
    "long-context-analysis": "claude-fable-5",
    "bulk-classification": "claude-haiku-4-5",
    "routine-edits": "claude-opus-4-8"
  }
}

This keeps Opus 4.8 as the cost-effective workhorse, uses Fable 5 only when its capabilities matter, and uses Haiku 4.5 for cheap high-volume work — surviving both June 22 credits and pay-as-you-go API economics.

Sources

Anthropic Newsroom: Claude Fable 5 and Claude Mythos 5 (June 9, 2026)
Anthropic developer documentation
claudefa.st: Claude Fable 5 Benchmarks, Pricing & June 22 Catch
llm-stats.com: Claude Fable 5 Review
agentpedia.codes: Claude Fable 5 Benchmarks and Prompting Guide
TrueFoundry: Claude Fable 5 API Benchmarks Pricing
Finout: Claude Fable 5 and Mythos 5 Pricing

Claude Fable 5 Rollout: Claude Code Pro/Max Credits June 22

The two changes

What Claude Fable 5 brings

What it costs

The June 22 credits change explained

Practical impact on Claude Code users

Light user (1–2 hours/day of Claude Code)

Heavy user (4+ hours/day, big agentic-coding workflows)

Pure API user (no claude.ai subscription)

How to manage cost — concrete tactics

1. Set explicit defaults and fallbacks

2. Pin models per workflow

3. Use the orchestrator-subagent pattern

4. Monitor burn rate

5. Cache aggressively

When to upgrade Pro → Max

When to downgrade Fable 5 → Opus 4.8

What’s likely next

Recommended setup for most developers as of June 10, 2026

Related reading

Sources