The cheapest Claude model is Claude Haiku 5.5, which costs $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, as of November 2026. Send it through the Batch API and the price halves, to $0.05 and $0.25.

Below is every model on Anthropic's API price list, ranked from cheapest to most expensive by standard price, checked on November 4, 2026. Prices are US dollars per million tokens.

The ranking at a glance

Rank Model Input Output Batch input / output
1 Haiku 5.5 (prompt up to 100K) $0.10 $0.50 $0.05 / $0.25
2 Haiku 4.5 $1 $5 $0.50 / $2.50
3 Sonnet 5.5 and Sonnet 5 $2 $10 $1 / $5
4 Sonnet 4.6 and 4.5 $3 $15 $1.50 / $7.50
5 Opus 5.5 $4 $20 $2 / $10
6 Opus 5 and Opus 4.5 to 4.8 $5 $25 $2.50 / $12.50
7 Fable 5.1 $10 $50 $5 / $25

One oddity stands out: the newest Sonnet and Opus cost less than older versions still on the list.

1. Claude Haiku 5.5

The clear price leader. Anthropic describes it as its fastest, most cost-efficient model and points it at high-volume work such as classification, extraction and routing. Watch the long-prompt tier: once a single request passes 100,000 tokens of input, the rate jumps to $0.50 in and $2.50 out.

2. Claude Haiku 4.5

Still listed, but ten times the price of Haiku 5.5 on short prompts. It also needs at least 4,096 tokens before a prompt can be cached, against 512 on Haiku 5.5. Unless you have a tested pipeline on it, Haiku 5.5 is the better buy.

3. Claude Sonnet 5.5 (tied with Sonnet 5)

The middle option and, for many teams, the best value. Anthropic calls Sonnet 5.5 its best mix of speed and intelligence. Sonnet 5 shares the same $2 and $10 price because a planned increase was cancelled.

4. Claude Sonnet 4.6 and 4.5

Older and pricier than Sonnet 5.5 at $3 and $15. There is little reason to choose them for new work.

5. Claude Opus 5.5

Anthropic's suggested starting model for most workloads and the daily driver for agentic coding. Its cache reads cost only $0.20, which softens the bill for long, repeated context.

6. Claude Opus 5 and Opus 4.5 to 4.8

All at $5 and $25, above the newer Opus 5.5.

7. Claude Fable 5.1

The top of the range at $10 and $50, aimed at long-running agents and the hardest reasoning. Its cache reads cost $0.25, a small slice of its input price.

The cheapest model is not always the cheapest bill

Two things change the real cost. Claude 4.7 and later models use a newer tokenizer that turns the same text into about 30 percent more tokens than older models, so compare bills, not only rates. And any model gets cheaper with the Batch API's 50 percent discount and cache reads, while US-only inference adds 10 percent and Opus 5.5 fast mode doubles the price.

Which model your coding tool actually burns

In Claude Code the model you pick drives the cost. Anthropic's cost guide says Sonnet handles most coding tasks well for less than Opus, and /model switches mid-session. To see which model your money really goes to, UsageAtlas reads your local Claude Code history and estimates cost per model. It is free, open source and runs on Windows, macOS and Linux. For apps you call the API from directly, the Claude Console is the place to compare models, since UsageAtlas tracks coding tools only.