Claude Opus 5 Pricing vs GPT-5.6, Grok 4.6, and Gemini 3.7
Opus 5 runs $5/$25 per million tokens, well above GPT-5.6, Grok 4.6, and Gemini 3.7 Flash. Here's the full price comparison and what it costs you.
Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens. That puts it well above OpenAI’s promotional GPT-5.6 price of $2 in / $10 out, xAI’s Grok 4.6 at $2 in / $6 out, and Google’s Gemini 3.7 Flash at 75 cents in / $3.75 out. Anthropic’s Fable 5 model runs even higher, at $10 in / $50 out. Sonnet 5, Anthropic’s mid-tier model, is more competitive at $2 in / $10 out, roughly matching GPT-5.6’s rate.
TL;DR
- Opus 5 is priced at a premium, $5/$25 per million tokens, against $2/$10 for GPT-5.6 and $2/$6 for Grok 4.6, meaning Anthropic is charging more per token for its flagship than its two biggest closed-model rivals.
- Gemini 3.7 Flash is the budget outlier, priced at 75 cents in and $3.75 out, undercutting every other model in this comparison by a wide margin.
- Token price is not the same as task cost, since independent testing from Code Rabbit found Opus 5 consumed roughly 50% more input tokens and 65% more output tokens than GPT-5.6 on identical review work, meaning the effective bill gap is larger than the sticker price suggests.
- Sonnet 5 is Anthropic’s competitive answer, priced at $2 in / $10 out, putting it on par with GPT-5.6 rather than asking customers to pay Opus-level rates for everyday work.
- Verbosity is driving up real bills, with developers describing Opus 5 as prone to turning small fixes into large rewrites, which matters more for total spend than the headline per-token rate.
- Anthropic’s own benchmarks look strong, including claims of more than doubling Opus 4.8 on a frontier coding benchmark, but pricing decisions are being made by developers testing actual workflows, not leaderboard scores alone.
- Market share data suggests price resistance is real, with reporting indicating Fable 5 attracted only 11% of tracked US customer spending after launch, while Opus 5 fared better commercially despite costing more.
How does Opus 5 pricing compare to GPT-5.6, Grok 4.6, and Gemini 3.7?
On raw per-token pricing, Opus 5 is the most expensive mainstream option among the four. Anthropic charges $5 per million input tokens and $25 per million output tokens for Opus 5. OpenAI’s GPT-5.6 promotional pricing sits at $2 in and $10 out, less than half of Opus 5’s rate on both ends. xAI’s Grok 4.6 comes in even lower on output at $2 in / $6 out. Google’s Gemini 3.7 Flash is the cheapest of the group by a wide margin, at 75 cents in and $3.75 out, aimed clearly at high-volume, cost-sensitive use cases rather than frontier reasoning tasks.
Anthropic’s Fable 5 model, positioned as a more restricted or specialized offering, costs even more than Opus 5: $10 in and $50 out. That’s five to 13 times Gemini 3.7 Flash’s rate depending on the token direction.
These numbers describe list price per token, not the cost of completing a given task. Because different models use different numbers of tokens to solve the same problem, the real-world spend comparison can look very different from the sticker price.
Why does token price alone not tell you what you’ll actually pay?
Per-token pricing assumes each model uses roughly the same number of tokens to do the same job. In practice, that assumption breaks down. Code Rabbit, which ran a controlled review comparing Opus 5 against a GPT-5.6-based production baseline, found that Opus 5 consumed about 50% more input tokens and roughly 65% more output tokens on the same review work. That means even where the per-token rate gap looks bad for Opus 5, the actual bill gap is worse once verbosity is factored in.
This lines up with qualitative complaints from developers. Theo Brown, the developer and CEO of T3 Chat, described Opus 5 treating minor comments like critical issues requiring thousands of lines of code. Code Rabbit’s testing also found Opus 5 caught fewer known problems than its production baseline while generating roughly four times as many low-value nitpicks. A model that writes more code to solve the same problem, or restructures more than necessary, drives up output token counts regardless of the advertised rate.
None of this means the underlying reasoning is weak. Opus 5 reportedly performs well on long autonomous builds, research tasks, computer use, and visually complex projects, areas where thoroughness pays off. The pricing problem shows up specifically in everyday, incremental work, the kind that made Claude popular with developers in the first place.
Is Opus 5 worth the premium?
That depends heavily on the workload. For long, autonomous, high-complexity tasks where Anthropic’s benchmark claims (including a reported doubling of Opus 4.8’s score on a frontier coding benchmark, and triple the next-best model on a separate reasoning benchmark) translate into real gains, the premium may be justified because fewer attempts are needed to reach a working solution.
Built like a system. Not vibe-coded.
Remy manages the project — every layer architected, not stitched together at the last second.
For routine, interactive coding work, incomplete instructions, follow-up questions, and simple fixes, the premium looks harder to defend. That’s the exact gap several experienced users have pointed to: strong benchmark performance paired with a frustrating day-to-day experience, verbose answers, and a tendency to over-engineer small requests. Benchmarks reward solving a well-defined task correctly. They don’t measure whether a model wastes tokens getting there or whether it correctly judges when a small fix should stay small.
Anthropic’s counterargument is that Opus can be efficient when it solves difficult problems in fewer attempts, and that Sonnet 5, at $2 in / $10 out, is available as the cost-competitive option for everyday work rather than asking every customer to pay Opus rates. That positioning makes some sense on paper. Sonnet 5’s pricing roughly matches GPT-5.6, so customers who don’t need Opus-level reasoning have a cheaper Anthropic option that doesn’t require switching providers.
How does Anthropic’s pricing strategy compare to its competitors’ market position?
OpenAI, xAI, and Google are all pricing more aggressively than Anthropic’s flagship, and the gap is not small. Grok 4.6’s output pricing of $6 per million tokens is less than a quarter of Opus 5’s $25. Gemini 3.7 Flash undercuts everyone, which fits Google’s apparent strategy of using Flash-tier models to win high-volume workloads while reserving other tiers for heavier reasoning tasks.
This pricing gap matters more now than it would have a year ago because switching costs have dropped. Claude Code built a reputation as the strongest serious coding assistant for much of the past year, but competitive alternatives have closed that gap. Financial Times reporting cited in industry commentary found Fable 5 attracted only 11% of tracked US customer spending after its launch, with price and the “good enough” performance of cheaper models cited as limiting factors, even though Opus 5 performed better commercially than Fable 5 did. That’s a meaningful signal: when a flagship model costs several times more than viable alternatives, customers increasingly test whether the premium is earned rather than assuming it by default.
Anthropic’s overall business hasn’t been hurt yet. Its annualized revenue run rate reportedly exceeded $65 billion by the end of July, up sharply from roughly $9 billion at the end of 2025, and the company has reportedly been preparing for a possible public offering. But revenue is a lagging indicator. If developers decide the premium isn’t earning its keep on daily work, that decision can show up in switching behavior well before it shows up in quarterly numbers.
Frequently Asked Questions
How much does Claude Opus 5 cost per million tokens?
Opus 5 costs $5 per million input tokens and $25 per million output tokens. Fable 5, a more restricted Anthropic model, costs more: $10 in and $50 out.
Is GPT-5.6 cheaper than Claude Opus 5?
Yes. OpenAI’s promotional GPT-5.6 pricing is $2 per million input tokens and $10 per million output tokens, less than half of Opus 5’s rate on both input and output.
What’s the cheapest model in this comparison?
Google’s Gemini 3.7 Flash, at 75 cents per million input tokens and $3.75 per million output tokens, is the cheapest option among Opus 5, GPT-5.6, Grok 4.6, and Gemini 3.7 Flash.
Does Opus 5 actually cost more in practice, not just per token?
One coffee. One working app.
You bring the idea. Remy manages the project.
Often yes. Code Rabbit’s testing found Opus 5 used about 50% more input tokens and 65% more output tokens than a GPT-5.6-based baseline on the same review work, meaning the real bill gap can exceed the listed per-token price difference.
Is Sonnet 5 a cheaper alternative to Opus 5?
Yes. Sonnet 5 is priced at $2 in / $10 out, matching GPT-5.6’s rate and positioned by Anthropic as the cost-competitive option for work that doesn’t require Opus-level reasoning.
