Opus 4.8 Still Owns 28% of Anthropic Spend

Ramp data shows Opus 4.8 at 28% spend vs Opus 5's 3.5%. Pin your Claude Code model to opus-4-8 for consistency, or use sonnet-4-6 for cheaper tasks.

Key Takeaways

• Ramp data shows Opus 4.8 at 28% spend vs Opus 5's 3.5%.
• Pin your Claude Code model to opus-4-8 for consistency, or use sonnet-4-6 for cheaper tasks.

The Data: What Developers Are Actually Paying For

Anthropic's July 2026 model spend data from the Ramp AI index — pulled from billing records of 70,000 companies — tells a story that contradicts the hype cycle. Opus 4.8, a model released months ago, holds 28.0% of all Anthropic spend. The brand-new Opus 5, released July 24th, captures just 3.5%.

Here's the full breakdown:

Model	% of Spend
Opus 4.8	28.0%
Sonnet 4.6	8.3%
Fable 5	8.0%
Opus 4.6	6.9%
Sonnet 5	3.6%
Opus 5	3.5%
Opus 4.7	1.7%
Sonnet 4.5	1.3%
Haiku 4.5	1.0%
Opus 4.5	0.7%

This isn't a lag effect. Opus 5 has been out for a full month, and developers have had time to evaluate it. The HN community sentiment explains why: "Opus 5 is not consistent. When it is better than 4.8, it's much better. But when it's worse than 4.8, it's much worse."

Why Consistency Beats Peak Performance in Claude Code

Claude Code is an agentic tool. It doesn't run a single prompt — it runs dozens, sometimes hundreds, of sequential calls to complete one task. Each call builds on the previous one's output. When a model is inconsistent, a single bad response in the middle of a multi-step workflow can cascade into a completely broken result.

Opus 4.8's advantage isn't that it's smarter. It's that it's predictable. Every call returns roughly the same quality. That predictability matters more than raw capability when you're chaining 50 tool calls together to refactor a codebase.

There's also a cost angle. Opus 5 likely carries a premium price tag, and the ROI isn't there when the marginal quality improvement is inconsistent. Developers are making a rational financial decision: pay less for a model that reliably gets the job done.

How To Apply This To Your Claude Code Setup

1. Pin Your Model in CLAUDE.md

If you're on Opus 5 and seeing inconsistent results, don't wait for Anthropic to fix it. Pin your model explicitly:

# In CLAUDE.md
## Model Configuration
- Use opus-4-8 for all code generation and refactoring tasks
- Use sonnet-4-6 for simple queries, file reads, and quick edits

2. Use the Model Flag Per-Session

For one-off tasks where you need maximum reasoning, you can still invoke Opus 5 selectively:

claude --model opus-5 "Design the database schema for a multi-tenant SaaS"

But for day-to-day work, default to 4.8:

claude --model opus-4-8 "Refactor this error handling into a middleware"

3. Match Model to Task Complexity

The spend data shows Sonnet 4.6 at 8.3% — developers are already using it for lighter work. Adopt this pattern explicitly:

• Simple tasks (formatting, docstrings, small edits): haiku-4-5 or sonnet-4-6
• Standard coding (implementing features, fixing bugs): opus-4-8
• Complex architecture (system design, tricky debugging): opus-5 but only when you can verify output

4. Set Up Model Fallbacks

Claude Code supports model fallbacks. Configure yours so that if Opus 5 fails or produces low-quality output, it automatically retries with 4.8:

claude --model opus-5 --fallback opus-4-8 "Solve this concurrency bug"

The Bigger Picture: Fable 5's Cautionary Tale

Fable 5's 8.0% share — despite being Anthropic's headline model — confirms that pricing drives adoption. The FT report notes Fable's cost made it less popular. The same economics apply to Opus 5. If you're paying per-token, you need the quality delta to justify the price delta. For most Claude Code workflows, it doesn't.

What This Means For Your Next Project

Before you start your next Claude Code session, ask yourself: am I using the right model for this task? The data says most developers should be on Opus 4.8 for heavy lifting and Sonnet 4.6 for lighter work. Opus 5 is a scalpel — use it when you need precision, not for everyday carving.

---

Source: simonwillison.net

[Updated 24 Aug via simon_willison]

New financial data from the FT, cited via people with knowledge of the matter, shows Anthropic's annualized revenue hit $65bn in July, up from $47bn in May. The company also told investors it expects Q3 to be profitable under the same model used to declare Q2 profitable, and counts 6,000 customers spending at least $100,000 annually. [per Simon Willison's FT coverage]

---

Originally published on gentic.news