Claude Fable 5.2 vs Claude Haiku 5.5
Claude Haiku 5.5 is 80.0x cheaper than Claude Fable 5.2 on our example workload, at list prices.
Side by side
| Measure | Claude Fable 5.2 | Claude Haiku 5.5 |
|---|---|---|
| Provider | Anthropic | Anthropic |
| Input / 1M tokens | $8.00 | $0.100 |
| Output / 1M tokens | $40.00 | $0.500 |
| Output price ÷ input price | 5.0x | 5.0x |
| Blended / 1M (3:1) | $16.00 | $0.200 |
| Context window | 1M | 1M |
| ≈ English words that fit | 750,000 | 750,000 |
| 1,000 calls (2000+500 tok) | $36.00 | $0.450 |
| AA Intelligence Index | 65 | 36 |
| LMArena Text | Not on LMArena yet | 1422 (high) |
| Blended $ per AA index point | $0.246 | $0.006 |
List prices from official pages (Anthropic, Anthropic), verified 2026-11-11. Benchmarks from public leaderboards (Artificial Analysis, LMArena), not run by us. Word counts use the rough 0.75 words per token rule for English. "$ per index point" is our own arithmetic (blended price ÷ index score), a rough value-for-money signal, not a benchmark.
What real workloads cost
Four example request shapes, priced at list rates with no caching or batch discounts. Monthly figures assume 100,000 requests.
| Workload | Claude Fable 5.2 | Claude Haiku 5.5 | Gap |
|---|---|---|---|
| Support chat reply a short system prompt, recent chat history and a reply | $0.0360 each $3,600.00 / mo | $0.00045 each $45.00 / mo | 80.0x |
| RAG answer a question plus about 20K tokens of retrieved documents, and a cited answer | $0.1920 each $19,200.00 / mo | $0.00240 each $240.00 / mo | 80.0x |
| 10-step agent task an agent loop of 10 model calls, each re-reading about 8K tokens of context and emitting a 400-token tool call | $0.8000 each $80,000.00 / mo | $0.0100 each $1,000.00 / mo | 80.0x |
| Long-report summary one 200K-token document summarized into about 2K tokens | $1.68 each $168,000.00 / mo | $0.0210 each $2,100.00 / mo | 80.0x |
When to pick each
- Pick Claude Haiku 5.5 when cost is the main constraint: at list prices it is 80.0x cheaper than Claude Fable 5.2 on a typical 2,000-in / 500-out call ($0.4500 vs $36.00 per 1,000 calls).
- Pick Claude Fable 5.2 when answer quality matters more than price: it scores 65 vs 36 on the Artificial Analysis Intelligence Index, and costs 80.0x more per call.
- Both come from Anthropic, so switching between them is usually a one-line model-name change on the same API and SDK; many teams route easy calls to the cheaper one and hard calls to the other.
Context and pricing limits
- Claude Fable 5.2: 1M tokens of context. Claude Haiku 5.5: 1M. Same size.
- Claude Haiku 5.5: Prompts up to 100K tokens; $0.50/$2.50 above. The prices above are this tier; check the official pricing page before committing.
- Cached input, batch and off-peak discounts can cut real bills a lot and differ by provider; they are not included here.
FAQ
Is Claude Haiku 5.5 cheaper than Claude Fable 5.2?
Yes, on our 2000-in / 500-out example call it costs $0.00045 vs $0.0360, 80.0x less.
Which has the larger context window?
Neither: both accept 1M tokens.
Which scores higher on benchmarks?
Claude Fable 5.2 scores 65 vs 36 on the Artificial Analysis Intelligence Index. On LMArena Text, Claude Fable 5.2 is Not on LMArena yet and Claude Haiku 5.5 is 1422 (high). Benchmarks are averages; test both on your own prompts.
Run your own numbers in the token cost estimator · Claude Fable 5.2 pricing · Claude Haiku 5.5 pricing · All comparisons