The Claude model family, compared
Where Claude Fable 5 sits relative to Opus 4.8, Opus 4.7, Sonnet 4.6, Haiku 4.5 - and its restricted sibling, Claude Mythos 5. All figures from Anthropic's official documentation, current as of June 10, 2026.
Model tiers
With the June 2026 launch, Anthropic's lineup spans four tiers. Mythos is a new class above Opus: Fable 5 is the Mythos-class model made safe for general release, while Mythos 5 itself remains gated behind Project Glasswing.
Claude Fable 5 & Claude Mythos 5
The frontier. Fable 5 (claude-fable-5) is generally available with real-time safety classifiers. Mythos 5 (claude-mythos-5) is the same model with cyber safeguards lifted, restricted to approved cyberdefense partners.
Claude Opus 4.8 & Opus 4.7
The workhorse frontier tier - state-of-the-art long-horizon agentic execution at half of Fable 5's token price. Opus 4.8 is the fallback model for Fable 5's safety classifiers.
Claude Sonnet 4.6
The best balance of speed, intelligence, and cost - 1M context, adaptive thinking, and 64K output at $3/$15 per million tokens.
Claude Haiku 4.5
Fastest and most cost-effective for simple, high-volume, latency-sensitive tasks - $1/$5 per million tokens.
Specifications
| Fable 5 | Opus 4.8 | Opus 4.7 | Sonnet 4.6 | Haiku 4.5 | |
|---|---|---|---|---|---|
| API model ID | claude-fable-5 | claude-opus-4-8 | claude-opus-4-7 | claude-sonnet-4-6 | claude-haiku-4-5 |
| Context window | 1M tokens | 1M tokens | 1M tokens | 1M tokens | 200K tokens |
| Max output | 128K | 128K | 128K | 64K | 64K |
| Input price /MTok | $10 | $5 | $5 | $3 | $1 |
| Output price /MTok | $50 | $25 | $25 | $15 | $5 |
| Batch price /MTok | $5 / $25 | $2.50 / $12.50 | $2.50 / $12.50 | $1.50 / $7.50 | $0.50 / $2.50 |
| Cache hit /MTok | $1 | $0.50 | $0.50 | $0.30 | $0.10 |
| Adaptive thinking | Always on | Opt-in | Opt-in | Opt-in | No |
| Extended thinking (budget_tokens) | No (400) | No (400) | No (400) | Deprecated | Yes |
| Effort levels | low-max | low-max | low-max | low-max | - |
| Reliable knowledge cutoff | Jan 2026 | Jan 2026 | Jan 2026 | Aug 2025 | Feb 2025 |
| Released | Jun 9, 2026 | 2026 | 2026 | 2025 | Oct 2025 |
Sources: Anthropic model overview and pricing documentation. Fable 5 prompt-cache writes: $12.50/MTok (5-minute TTL), $20/MTok (1-hour TTL); minimum cacheable prefix 512 tokens.
Benchmark deltas: Fable 5 vs Opus 4.8
| Benchmark | Fable 5 | Opus 4.8 | Delta |
|---|---|---|---|
| SWE-Bench Pro | 80.3% | 69.2% | +11.1 pts |
| FrontierCode Diamond | 29.3% | 13.4% | 2.2x |
| Terminal-Bench 2.1 | 88.0% | 82.7% | +5.3 pts |
| Humanity's Last Exam (no tools) | 59.0% | 49.8% | +9.2 pts |
| GDPval-AA (Elo) | 1932 | 1890 | +42 |
| OSWorld-Verified | 85.0% | 83.4% | +1.6 pts |
| ExploitBench | 78.0% | 40.0% | ~2x |
| HealthBench Professional | 66.0% | 56.9% | +9.1 pts |
Full benchmark analysis including GPT-5.5 and Gemini 3.1 Pro ->
Which model should you choose?
Choose Fable 5 when...
The task is genuinely hard: multi-day agentic runs, codebase-scale migrations, frontier research, complex legal or financial analysis, or anything where a failed attempt costs more than the tokens. Early adopters report it solves in one shot what Opus needs several attempts for - often using about half the tokens, which can bring effective cost close to parity despite the 2x price.
Choose Opus 4.8 when...
You need frontier-grade agentic coding at predictable cost, your workload is high-volume, or you operate under zero-data-retention requirements (Fable 5 is a "Covered Model" with mandatory 30-day retention). It remains Anthropic's recommended default for most production work.
Choose Sonnet 4.6 when...
Speed and cost matter as much as intelligence: interactive products, high-throughput pipelines, and most day-to-day coding. Same 1M context window at 30% of Fable 5's input price.
Choose Haiku 4.5 when...
The task is simple and latency-critical: classification, extraction, routing, autocomplete, or sub-agent fan-out where each call is small and the volume is huge.
And Claude Mythos 5?
Mythos 5 is the same underlying model as Fable 5 with cyber safeguards lifted - "the strongest cybersecurity capabilities of any model in the world." It is available only to approved Project Glasswing partners: cyberdefenders, critical-infrastructure operators, and government. Its predecessor, Claude Mythos Preview, found more than 10,000 high- and critical-severity zero-day vulnerabilities. Read the full Mythos 5 explainer ->
Ready to build with Fable 5?
Start with our API quickstart - model ID, breaking changes, effort tuning, and working code in five minutes.
Read the quickstart