Fable 5.2 Cost per Task: Cheaper Tokens, a 31% Higher Bill
Fable 5.2 cut list prices 20% to $8 input and $40 output per 1M tokens, yet SemiAnalysis measured its cost per task rising 31%, from $67.16 to $88.15, because the model uses far more tokens to check its own work.
Last updated: November 2026
Key takeaways
- Fable 5.2 lists at $8/$40 per 1M tokens versus $10/$50 for Fable 5.1, a 20% cut.
- SemiAnalysis's GDPx cost per task rose 31%, from $67.16 to $88.15.
- The extra cost came from behavior: Fable 5.2 sometimes wrote hundreds or thousands of test cases and spun up subagents, where earlier models wrote about 20.
- Fable 5.2 leads the Artificial Analysis Intelligence Index at 65, ahead of Gemini 4 Carbon (60) and Opus 5.5 (59).
- Forecast by cost per successful task, not by list price.
Why did cost per task rise when prices fell?
Price per token times tokens used equals the bill. Fable 5.2 lowered the first number and raised the second by more. According to Latent Space's AINews recap, earlier models tried about 20 test cases, while Fable 5.2 sometimes wrote hundreds or thousands and used subagents to verify its work. That is better engineering and a bigger invoice.
A 20% price cut is covered by a 25% rise in tokens, since 0.8 x 1.25 = 1.0. The measured 31% rise in cost per task implies token use grew by roughly 64% on that benchmark (1.31 / 0.8 = 1.64), an illustrative back-of-envelope figure that assumes the same input-to-output mix.
How do the models compare on intelligence and price?
| Model | Intelligence Index | Input $/1M | Output $/1M | |---|---|---|---| | Fable 5.2 (max) | 65 | $8 | $40 | | Gemini 4 Carbon (high) | 60 | not stated | not stated | | Opus 5.5 (max) | 59 | $4 | $20 | | GPT-6 Astra (max) | 57 | not stated | not stated | | Fable 5.1 (max) | 49 | $10 | $50 |
Index scores are from Artificial Analysis as reported by Latent Space; Opus 5.5 prices are from TrueFoundry's table. On Artificial Analysis, Fable 5.2 at high or xhigh effort used about 98k output tokens per task versus 173k for Opus 5.5 at max. Output-only, that is roughly $3.92 for Fable 5.2 (98k x $40 per 1M) versus $3.46 for Opus 5.5 (173k x $20 per 1M), illustrative and excluding input. Cost per task depends heavily on the benchmark, which is exactly why you must measure on your own tasks.
How do you calculate cost per task?
Use this formula: cost per task = (uncached input tokens x input rate + cached input tokens x cache rate + output tokens x output rate) x steps per task, then divide by the share of tasks that succeed to get cost per successful task. Our explainer on cost per token vs cost per task walks through it, and tokenizer inflation shows why token counts differ by model.
What should founders do before switching to Fable 5.2?
- 1Run 50 to 100 real tasks on Fable 5.2 and your current model, and log total tokens.
- 2Cap reasoning effort and subagent fan-out where the task allows.
- 3Reprice or add an effort tier if your plan is flat-rate, since heavier checking raises your cost per user.
- 4Compare providers on the Anthropic pricing page and plug results into the Calcaas LLM cost calculator.
For agent-style products, also see the AI agent use case calculator.
Frequently asked questions
Is Fable 5.2 cheaper than Fable 5.1?
Per token, yes: $8 and $40 versus $10 and $50 per 1M tokens. Per task, SemiAnalysis measured a 31% increase, from $67.16 to $88.15.
Why does Fable 5.2 use more tokens?
It verifies its work more aggressively, sometimes writing hundreds or thousands of tests and using subagents, where earlier models wrote about 20.
What does cost per task mean?
It is the total spend to complete one unit of work, including input, cached input, output and retries. Dividing by the success rate gives cost per successful task.
Is Fable 5.2 the best model?
It ranks first on the Artificial Analysis Intelligence Index with a score of 65, ahead of Gemini 4 Carbon at 60 and Opus 5.5 at 59, per Latent Space's report. The takeaway: a price cut is not a bill cut, so benchmark tokens per task before you migrate. Test your own scenario in the Calcaas LLM cost calculator. Note: place the JSON-LD above inside a <script type="application/ld+json"> tag in the page head.
More from the blog
LLM EconomicsFixed-Price vs Pay-As-You-Go Inference: Breakeven Utilization
Fixed-price inference plans beat pay-as-you-go only once your traffic fills roughly 60 to 80% of the monthly token allowance; the breakeven utilization is u = P / (A x r), the plan price divided by what the same allowance would cost metered.
LLM EconomicsClaude API Pricing per Million Tokens: Every Model, Cache and Batch Rate
Claude API pricing runs from $0.10 per 1M input tokens on Haiku 5.5 to $8 on Fable 5.2, with output always about 5x the input price, cache reads at a fraction of input, and a 50% discount for batch jobs returned within 24 hours.
LLM EconomicsClaude Prompt Caching Break-Even: The Math Behind the Savings
Claude prompt caching breaks even on the second request: a cache write costs 1.25x the base input price, a read costs 0.1x (0.05x on Opus 5.5 and Sonnet 5.5), so a single reuse inside the 5-minute window repays the write premium.
Pricing math, in your inbox.
One short note a week on AI pricing, token economics, and margin. No spam, unsubscribe anytime.