Claude Sonnet 5 Just Shipped at $2/$10 Pricing. Long-Context Reasoning at 90% of Opus. Here's How Your Consulting Rate Just Shifted.
On July 8, Anthropic released Claude Sonnet 5 with introductory pricing of $2 per million input tokens and $10 per million output tokens (regular pricing $3/$15 after August 31). The model handles 200K token contexts, performs within 5-10% of Opus on most benchmarks, and matches Opus on long-context reasoning tasks.
For anyone selling AI consulting services, this is the moment your hourly rate became less valuable than it was last quarter.
The unit economics
A 10,000 token reasoning task costs:
- Claude Opus: $0.045 (at $4.50/$45 pricing)
- Claude Sonnet 5: $0.02 (at $2/$10 pricing)
- Your consulting billable: $200-400 per hour
If it takes you 30 minutes to do that task (setting up prompts, validating output, writing the report), you're billing $100-200. The model did the reasoning for $0.02.
The gap widens if you're doing higher-leverage work, say, reviewing and integrating 50 documents into a knowledge base. That's a $0.02 per document model cost, or $1 total, versus your $200-400 daily rate for the same work.
This isn't new. Claude 3 Opus already made some consultant labor cheaper than hiring. Sonnet 5 just moves the line.
What tasks got commoditized
Pure reasoning work:
- Document summarization and analysis
- Code review and refactoring suggestions
- Strategic recommendations (if you're just synthesizing existing data)
- Writing and editing (first draft)
- Data extraction and transformation (if the structure is clear)
- Research synthesis (if the sources are clear)
These are all now model-bounded, not labor-bounded. You can't charge $200/hour for work the model does in 2 seconds for $0.02.
The work that didn't get cheaper:
- Understanding what to ask the model
- Validating output in context (does this recommendation make sense for this customer?)
- Connecting the output to business outcomes
- Handling edge cases and exceptions
- Reframing the problem when the first approach doesn't work
The market signal
Anthropic's pricing strategy is clear: they want teams to treat reasoning as a utility. Pay for tokens, not for reasoning as a service. If you're still charging for pure reasoning as a consultant, you're fighting the market.
Here's what that looks like on your P&L if you're a small agency:
Before (June 2026):
- Document analysis: $200/day (8 hours) × 20 days/month = $4,000
- Model cost (Claude Opus): $50/month
- Margin: $3,950/month
After (August 2026 onward):
- Document analysis: same rate, but clients now know Claude Sonnet 5 can do it for $1/day
- You're bidding against $1/day, not against another consultant's $200/day
- Your margin compresses to $3,800/month... until the client notices and asks why you're not passing through the savings
The honest take: your advantage is context, not reasoning
The consultants who thrive won't be the ones running the model. They'll be the ones who understand the customer's business deeply enough to know:
- What questions to ask the model
- Which outputs are worth acting on and which are hallucinations
- How to connect model outputs to business metrics
- What to do when the model says "I don't know" (which matters in real customer scenarios)
A fintech consultant who knows the regulatory landscape around loan origination can use Claude Sonnet 5 to speed up her work. A generalist who just runs documents through Claude competes on cost, not expertise.
The fee shift is from "I did the reasoning for you" to "I know what reasoning matters for your business."
Where pricing power remains
Domain expertise: If you understand pharmaceutical supply chain optimization, you can use Sonnet 5 to move faster, but you still charge domain rates. The model is your tool, not the service.
Validation and accountability: "Claude says this" is free information. "I've validated this against your data and I'm willing to bet on it" is a service. If a customer is making a $500K decision, they're not paying for the reasoning. They're paying for your judgment.
Integration and operations: Connecting Claude to your customer's systems, maintaining the pipeline, handling exceptions, retraining models when performance drifts: these are services, not reasoning.
Strategic framing: The work you do before you ever touch Claude. "Here's what we're actually trying to optimize for" is worth more than the reasoning that follows.
What I'd actually do
If you're a consultant making money from AI:
Audit your billable hours by category. How much of your time is pure reasoning (easy to commoditize) vs. context (hard to commoditize)? Aim for >60% context work.
Repackage pricing. Stop charging hourly for reasoning tasks. Charge project-based fees for "analyze this document set and give me 5 recommendations" (the output matters, not how long Claude took). Charge for your domain expertise, judgment, and integration work.
Compete on speed and integration, not cost. You're not competing on "I can do this reasoning" (so can Claude). You're competing on "I can connect this to your business and make it operational in 2 weeks."
Price your time accordingly. If you're spending 50% of your engagement time validating and integrating Claude's output, and 50% reasoning, you've just doubled the value you provide per reasoning task. Price it that way.
For solo operators: the transition is painful. Your highest-volume, most-scalable service (reasoning-heavy work) just got commoditized. If you've built a business on that, you have a 6-month window to reposition. Either add more context-based work (domain expertise, validation, integration) or repackage as "AI implementation" (you bring the domain knowledge, Claude brings the reasoning).
The market for pure reasoning consultation is shrinking. The market for domain-expert-plus-Claude is growing. Pick which side of that line you're on.
Author
Lukas
@lukcombinator