Table of Contents
Claude Opus 5 vs Fable 5: What Half the Price Means for Families
Claude Opus 5 vs Fable 5 in plain math: $5/$25 against $10/$50 per million tokens, what a homework session costs, and how to teach kids to read an AI bill.
A single homework question to Claude Opus 5 costs about a cent and a half. The same question to Claude Fable 5 costs about three cents. That is the whole Claude Opus 5 vs Fable 5 price story in one sentence, and it is also the reason “half the price” is a more useful phrase for developers than for families, who pay a flat $20 a month and never see a token.
Still, the per-token math is worth learning, because every tutoring app, school platform, and homework tool your kid touches is paying it and passing it along. Here is what the numbers are, where they come from, and how to turn them into a lesson.
Key Takeaways
- Claude Opus 5 (July 24, 2026) is priced at $5 per million input tokens and $25 per million output tokens; Fable 5 and Fable 5.1 are $10 and $50, exactly double.
- At those rates, a 150-word homework question with a 350-word answer costs roughly 1.4 cents on Opus 5 and 2.9 cents on Fable 5, before reasoning tokens.
- Fable 5.1 (September 1) cut its cache-read price from $1.00 to $0.25 per million tokens, which narrows the gap on long tutoring sessions where the same conversation is resent every turn.
- Anthropic’s own docs tell developers to “start with Claude Opus 5” and reserve Fable for demanding, long-running work; that advice applies doubly to homework.
- A Claude Pro subscription is $20 a month with Opus included; Fable access on Pro and Max plans draws down usage credits at 50% of weekly limits.
Claude Opus 5 vs Fable 5: what each model actually is
Claude Fable 5 is Anthropic’s most capable widely available model, released June 9, 2026, with a 1-million-token context window and safety classifiers that decline certain cybersecurity, biology, and model-copying requests. Claude Opus 5, released July 24, is a separate model that Anthropic says “comes close to the frontier intelligence of Claude Fable 5 at half the price.” Fable 5.1 arrived September 1 at the same per-token prices as Fable 5.
The company’s own comparisons are specific. On CursorBench 3.2 at maximum effort, Opus 5 lands “within 0.5% of Fable 5’s peak score, but at half the cost per task.” On OSWorld 2.0, a computer-use benchmark, Anthropic reports Opus 5 surpassing Fable 5’s best result “at just over a third of the cost.” Those are vendor numbers run in the vendor’s harness, so treat them as directional. The direction, though, is clear: for most tasks, Anthropic itself does not think you need Fable.
Where Fable keeps its edge is long, multi-hour agent work and the strictest safeguards. Its cyber classifiers, per Anthropic, intervene about 85% less often on Opus 5 than on Fable 5, which is one reason the cheaper model can feel less fussy in day-to-day use.
How per-token pricing turns into a homework bill
A token is a chunk of text, usually a word or part of one. A rough rule for English is 1.3 tokens per word, but Anthropic’s token-counting documentation notes that Fable 5, Fable 5.1, and the Opus 4.7-and-later family use a tokenizer that produces roughly 30% more tokens for the same text than older Claude models. The estimates below use 1.5 tokens per word to stay conservative.
Prices are quoted per million tokens, split into input (what you send, including the whole conversation so far) and output (what the model writes back). Per the Fable 5 platform docs, Fable 5 is $10 input and $50 output. Opus 5 is $5 and $25. Two things make real bills higher than the naive math:
- Reasoning tokens. Adaptive thinking is always on for these models, and the tokens the model spends thinking are billed as output. A hard math problem at high effort can quietly triple the output count.
- Resent context. Every turn of a chat resends the entire conversation as input. A 15-turn tutoring session does not cost 15 questions; it costs 15 questions plus 14 copies of the growing history.
That second point is exactly what prompt caching addresses. The Fable 5.1 release notes cut cache reads from $1.00 to $0.25 per million tokens, so a resent, unchanged conversation prefix costs a fortieth of fresh input. We cover the mechanism in our prompt caching explainer; the takeaway here is that long sessions on Fable 5.1 are closer to Opus 5 pricing than the headline “double” suggests.
How to Teach Your Kid About AI Pricing
Ages 5–8: Word coins
Put 20 coins in a jar and explain that the “robot helper” charges one coin for every sentence it says back. Ask it (or play the robot yourself) a question and count the sentences in the answer, then pay. Kids discover fast that “tell me everything about volcanoes” empties the jar, while “what makes a volcano erupt?” costs three coins. Asking a precise question is the skill.
Ages 9–12: The homework receipt
Have your kid write down one real homework question and count its words. Multiply by 1.5 to estimate tokens. Then guess the answer length in words and do the same. Using a calculator and the prices above, work out the Opus 5 cost and the Fable 5 cost. Most kids are startled that a question costs less than a penny and delighted that they can now explain why to an adult.
Ages 13+: Build the family AI budget
For one week, your teen logs every AI session: tool, turns, rough length. At the end, they estimate the API cost at Opus 5 and Fable 5 rates, then compare it with a $20 Claude Pro plan or whatever your family pays. They present a one-page recommendation: subscription or pay-per-use, and which model tier for which tasks. Bonus challenge: find the cache-read price on Anthropic’s pricing page and explain to a parent why it matters for long chats.
The question to ask: “If Fable costs twice as much per token, why might a long tutoring session on Fable 5.1 cost less than twice as much as the same session on Opus 5?”
Cost per task: Opus 5, Fable 5, and Fable 5.1 with caching
Estimates assume 1.5 tokens per word, list prices, and no reasoning tokens. Real costs run higher with high effort settings.
| Task | Tokens (in / out) | Opus 5 | Fable 5 | Fable 5.1 with caching |
|---|---|---|---|---|
| One homework question (150-word prompt, 350-word answer) | 225 / 525 | ~1.4¢ | ~2.9¢ | ~2.9¢ (nothing to cache) |
| Essay feedback (800-word essay, 500-word critique) | 1,200 / 750 | ~2.5¢ | ~5.0¢ | ~5.0¢ |
| 45-minute tutoring chat (15 turns, history resent) | ~60,000 / ~8,000 | ~$0.50 | ~$1.00 | ~$0.65–0.70 |
| A month of tutoring (20 sessions) | ~$10 | ~$20 | ~$13–14 | |
| Claude Pro subscription (flat, Opus included) | $20/month | Fable draws usage credits |
Read the bottom two rows together. A kid doing 20 tutoring sessions a month on Opus 5 through the API would spend about $10; on a Pro plan they would spend $20 and get everything else in the plan. For most families the subscription is the real price, and the API math is a way to understand what the subscription is buying.
What the research says about whether the expensive model helps learning
There is no study comparing Opus 5 and Fable 5 in a classroom, and anyone who tells you the pricier model teaches better is guessing. What we do have is evidence about what kind of AI use helps.
The strongest single trial is Kestin et al. (2025) in Scientific Reports, a randomized crossover study of 194 Harvard physics students. Students working with a carefully designed AI tutor learned more than students in an active-learning classroom, with median gains more than double the control condition. The important detail for our purposes: the tutor was built with expert-authored scaffolds and guardrails that forced step-by-step reasoning. The design did the work, not the size of the model underneath.
On the other side, the OECD’s PISA 2025 results (September 8, 2026) found that 15-year-olds who used AI daily to draft written assignments scored 28 points lower in science than those who never did, after adjusting for socioeconomic status, while students who used a chatbot about weekly to help them learn posted the highest scores of any group. Associations, not causation, but the pattern is consistent: how the tool is used matters more than which tier you paid for.
Common Sense Media’s August 18, 2026 survey of 1,017 teens adds that 63% of teens who use AI for schoolwork get an answer from it, and only 27% had a teacher discuss safe use. A cheaper model that explains is worth more than an expensive one that answers.
What to actually do at home
Default to Opus 5 and forget about Fable
Anthropic’s own guidance is to start with Opus 5. For homework, that is the end of the decision. If your kid hits a task where Opus 5 fails and Fable succeeds, you will know, and it will be rare.
Turn the receipt into a habit
Once a month, have your kid estimate what their AI use would have cost at API rates. It takes five minutes and keeps the abstraction of “AI is free” from taking root. Our tokens and per-million pricing explainer has a worksheet version.
Set the effort dial low by default
Where the app exposes an effort or reasoning setting, teach your teen to leave it low for routine work and raise it only for a problem they have already tried twice. Reasoning tokens are billed as output, and “max effort on everything” is how a $20 plan runs out of usage by Wednesday.
Audit the subscriptions, not the tokens
The family bill is driven by plans, not per-question pricing. Count how many AI subscriptions your household holds across ChatGPT, Claude, Gemini, and school apps. Our piece on subscription creep has the checklist. For context on what each summer 2026 model costs, see the parents’ guide to summer 2026 models.
What not to do
Do not buy a Max plan for Fable access “so the kid has the best.” At $100 a month and up, it is priced for professionals running hours-long agents. A homework question does not get better with a bigger model; it gets better with a better question.
What to Watch For Over the Next 3 Months
- Week 4: Check which model your kid’s Claude account defaults to. If it is Fable and they are on Pro, usage credits are draining at 50% of weekly limits for tasks Opus 5 handles fine.
- Month 2 red flags: “Usage limit reached” messages mid-week; a teen insisting only the expensive model “understands” them; app subscriptions that quietly moved to a higher tier.
- Month 3 self-check: Re-run the homework receipt with your kid. If their estimate is within a factor of two of the real per-task cost, they understand the pricing model better than most adults.
Frequently Asked Questions
Is Claude Opus 5 really half the price of Fable 5?
Per token, yes: $5/$25 versus $10/$50 per million input and output tokens. Per task, it depends. Anthropic reports Opus 5 within 0.5% of Fable 5 on one coding benchmark at half the cost, and on a computer-use benchmark at about a third. Long sessions on Fable 5.1 narrow the gap thanks to cheaper cache reads.
How much does one AI homework question cost?
Using list prices and no reasoning tokens, a 150-word question with a 350-word answer costs about 1.4 cents on Opus 5 and 2.9 cents on Fable 5. Reasoning tokens and longer conversations push real costs higher, sometimes several times higher at high effort.
Should my kid use Fable 5 for schoolwork?
Probably not. Anthropic’s documentation tells developers to start with Opus 5 and use Fable for demanding, long-horizon work. Homework is neither. Fable also carries stricter classifiers and 30-day data retention, which matter more for enterprises than for a ninth grader.
Does a Claude Pro plan include Fable?
Pro is $20 a month and includes Opus. Fable is available on Pro and Max plans through usage credits at 50% of weekly limits, according to Anthropic’s pricing page. In practice, a Pro plan set to Opus 5 covers ordinary study use.
What is a cache read and why is it cheaper?
When a chat resends the same conversation history each turn, the model can reuse a stored version of that prefix instead of reprocessing it. Fable 5.1 charges $0.25 per million tokens for those cache reads versus $10 for fresh input. Long tutoring chats benefit the most.
About the author
Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.
Sources
- Anthropic. (2026, July 24). “Introducing Claude Opus 5.” Anthropic News. https://www.anthropic.com/news/claude-opus-5
- Anthropic. (2026, June 9). “Introducing Claude Fable 5 and Claude Mythos 5.” Claude Platform Documentation. https://platform.claude.com/docs/en/models/fable-5/introducing-claude-fable-5-and-claude-mythos-5
- Anthropic. (2026, September 1). “What’s new in Claude Fable 5.1.” Claude Platform Documentation. https://platform.claude.com/docs/en/models/fable-5-1/whats-new-fable-5-1
- Anthropic. (2026). “Token counting.” Claude Platform Documentation. https://platform.claude.com/docs/en/build-with-claude/token-counting
- Anthropic. (2026). “Pricing.” https://claude.com/pricing
- Kestin, G., Miller, K., Klales, A., et al. (2025). “AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting.” Scientific Reports. https://www.nature.com/articles/s41598-025-97652-6
- OECD. (2026, September 8). “PISA 2025 Results (Volume I).” https://www.oecd.org/en/publications/pisa-2025-results-volume-i_73451bc5-en.html
- Common Sense Media. (2026, August 18). “Teens in the AI Era: Schoolwork and the Skills That Matter.” https://www.commonsensemedia.org/research/teens-in-the-ai-era-schoolwork-and-skills-that-matter
