In September 2026, Anthropic released three models in three weeks: Fable 5.1 (September 1), Opus 5.5 (September 22), and Sonnet 5.5 (September 28). For businesses already using Claude, the relevant part is what changes in practice and when it’s worth migrating.
Claude timeline in 2026
| Model | Date | Input/M | Output/M | Context | Key note |
|---|---|---|---|---|---|
| Opus 4.5 | Jan 2026 | ~$15 | ~$75 | 200K | First 4.x version |
| Opus 4.6 | Feb 2026 | ~$15 | ~$75 | 1M | First model with 1M context |
| Mythos Preview | Apr 2026 | Restricted | Restricted | — | Invite-only (Project Glasswing) |
| Opus 4.7 | Apr 2026 | ~$15 | ~$75 | 1M | Better vision, better code |
| Fable 5 | Jun 2026 | ~$12 | ~$60 | 1M | Public “Mythos-class” |
| Fable 5.1 | Sep 1 | $10 | $50 | 1M | 25% cheaper than Fable 5 |
| Opus 5.5 | Sep 22 | $4 | $20 | 1M | Fable-level at 60% less |
| Sonnet 5.5 | Sep 28 | $2 | $10 | 1M | 30% faster than Sonnet 5 |
Current family: Fable 5.1 (premium), Opus 5.5 (flagship), Sonnet 5.5 (balanced), Haiku 4.5 (economy). All with 1M context.
Fable 5.1: maximum capability, fewer false positives
Fable 5.1 is Anthropic’s most capable model, designed for long-running tasks spanning hours and multiple applications.
What changed from Fable 5:
- 60% fewer security false positives: Claude Code users reported Fable 5 rejecting legitimate operations out of excessive caution. Fable 5.1 corrects this
- Invisible text watermarking: Fable 5.1 includes invisible watermarks in generated text. Anthropic is rolling out a detection API in private preview
- 25% cheaper: Significantly reduced cache reads, lowering real cost 25% on typical workloads and up to 45% on cache-heavy agentic work
Opus 5.5: Fable level at 40% less
Opus 5.5 matches Fable 5.1 on most benchmarks and practical tasks, at $4/$20 versus $10/$50. The performance difference shows in:
- Very long-horizon tasks (>4 hours continuous execution)
- Extreme reasoning where Fable has a slight edge
- Cybersecurity work where Fable was evaluated with stricter standards
For 90% of enterprise tasks (code, document analysis, agents, support), Opus 5.5 and Fable 5.1 are indistinguishable in quality.
Concrete numbers
- Solves more command-line tasks than Opus 5 making 40% fewer calls
- Uses half the tokens of Opus 5 for the same result
- Cache reads at $0.20/M tokens
Sonnet 5.5: the production model
Sonnet 5.5 (September 28) generates output 30% faster than Sonnet 5 at the same price: $2/$10. For most businesses, this is the model to use as baseline.
- Same price as GPT-6 Sol ($2/$10)
- 1M context (vs Sol’s 272K)
- 128K max output (vs Sol’s 32K)
- Native MCP
How to choose between Fable, Opus and Sonnet
| Criteria | Sonnet 5.5 | Opus 5.5 | Fable 5.1 |
|---|---|---|---|
| Price (input/output M tokens) | $2/$10 | $4/$20 | $10/$50 |
| Routine tasks | Sufficient | Excellent | Overkill |
| Complex reasoning | Good | Excellent | Excellent |
| Long agentic work | Good | Very good | Excellent |
| Code and debugging | Very good | Excellent | Excellent |
The rule of thumb: start with Sonnet 5.5. Step up to Opus 5.5 for tasks where Sonnet doesn’t reach. Use Fable 5.1 only when the marginal quality difference justifies 2.5x the cost.
We built our own website with Claude Code, and the experience with different Opus versions is documented in how we built the web with Claude Code. The MCP protocol created by Anthropic gives Claude a technical advantage in agentic architectures.
Our team works with Claude API since the earliest models. If you need to design a multi-tier architecture or migrate from Opus 4.x to 5.5, we can help.
Request a free consultation and we’ll analyze your case.