Every Claude model now costs 25% of Anthropic's list price
Since October 19, 2026, Claude Opus 5.5, Sonnet 5.5, Haiku 5.5 and Fable 5.1 cost 75% less on QuickSilver Pro than at Anthropic's list price, cache reads and writes included. Nothing to change in your code.
On October 19, 2026 we cut the price of every Claude model on QuickSilver Pro to a quarter of Anthropic's list price. Before that, the discount was 50–60% depending on the model. The new prices apply to all accounts automatically; there is no plan to choose and no code to change.
New prices
Per million tokens, input / output:
| Model | Anthropic list | QuickSilver Pro | You save |
|---|---|---|---|
| Claude Opus 5.5 | $4.00 / $20.00 | $1.00 / $5.00 | 75% |
| Claude Fable 5.1 | $10.00 / $50.00 | $2.50 / $12.50 | 75% |
| Claude Sonnet 5.5 | $2.00 / $10.00 | $0.50 / $2.50 | 75% |
| Claude Haiku 5.5 | $0.10 / $0.50 | $0.025 / $0.125 | 75% |
The older Claude models (Opus 5, Fable 5, Opus 4.8, Sonnet 5 and Haiku 4.5) moved to the same rate; their prices are in Claude API pricing in 2026.
Three details that matter for real bills:
- Prompt caching is discounted the same way. Cache writes and cache reads are also 25% of Anthropic's list price, so a request that reuses a cached prompt prefix keeps the full discount on top of the caching saving.
- Haiku 5.5's long-prompt rate is discounted too. A request whose prompt is longer than 200K tokens is billed at $0.125 / $0.625 per million tokens.
- Billing is still prepaid and per request. Every request appears in your dashboard with its model, token counts and cost, so you can check the new rate against your own traffic.
What it means in practice
Output tokens are where Claude bills add up, and the same ratio applies there: whichever Claude model a session runs on, every input, output and cache token in it is 75% cheaper than at Anthropic's list price.
For coding agents the difference is larger than it looks, because tools such as Claude Code resend a long system prompt and tool list with every request. We measured a small bug fix in Claude Code: three requests, about 62,000 input tokens, mostly served from the prompt cache. The setup and the numbers are in Claude Code with QuickSilver Pro.
If you are already using QuickSilver Pro
Nothing to do. Your API key, base URL and model IDs stay the same, and requests made since the change are billed at the new rate.
If you are not
Create a key in the dashboard and point your client at QuickSilver Pro. With the Anthropic SDK that is one argument:
import anthropic
client = anthropic.Anthropic(
base_url="https://api.quicksilverpro.io",
api_key="YOUR_QSP_KEY",
)
message = client.messages.create(
model="claude-sonnet-5-5",
max_tokens=1024,
messages=[{"role": "user", "content": "Summarise this pull request: ..."}],
)
print(message.content[0].text)
New accounts get $0.05 of free credit, which covers the starter models (GPT-6 Luna, Claude Haiku 5.5, DeepSeek V4.1 Flash, Qwen3.8 Flash Next and MiMo-V2.6-Flash). The larger Claude models need a top-up, from $5.