Anthropic quadruples the price of its new Haiku model

AI developers have gotten used to a predictable script: newer models get smarter, faster, and aggressively cheaper. Anthropic just flipped that script on its head.

Anthropic officially launched Claude 3.5 Haiku through its API and third-party platforms like AWS Bedrock. But instead of replacing its predecessor at the same budget tier, the company rolled out a massive 4x price hike.

Smarter than Opus, but quadrupled in price

When Anthropic first teased Claude 3.5 Haiku last month, the expectation was that pricing would stay flat compared to Claude 3 Haiku. Now, the company has officially changed its tune.

In a post on X, Anthropic explained that during final testing, the new Haiku surpassed Claude 3 Opus, its former heavyweight flagship, across several benchmarks at a fraction of the cost. To reflect that leap in intelligence, Anthropic decided the price tag needed to go up.

Claude 3.5 Haiku now costs $1 per million input tokens and $5 per million output tokens. That compares to $0.25 per million input tokens and $1.25 per million output tokens on Claude 3 Haiku, representing a clean quadrupling of costs across the board.

Anthropic points out that the model shines at coding assistance, data extraction, labeling, and content moderation. It also features a more recent knowledge cutoff and can generate longer chunks of text.

Higher costs with missing eyesight

The price hike is tough enough, but early adopters will also notice an unexpected downgrade: Claude 3.5 Haiku cannot analyze images at launch.

That puts it behind Claude 3 Haiku in multimodal tasks. Alex Albert, head of developer relations at Anthropic, confirmed on X that the older model will remain available for teams that need maximum cost-efficiency and image processing.

While the reasoning makes technical sense from a raw benchmark perspective, an AI lab jacking up prices on a product lineup is virtually unheard of in this current cycle.

Why it matters

In an industry where everyone from OpenAI to Google is racing to the bottom on token pricing, Anthropic's move feels daring, if not slightly disconnected from developer expectations.

For builders, Haiku was never meant to be a luxury model. It was the default workhorse for bulk scraping, classification, and background automations. A 4x price jump completely upends unit economics for high-volume pipelines, and we suspect many developers will either stay parked on Claude 3 Haiku or migrate to competing micro-models that still offer vision and cheaper rates.

We think Anthropic's logic is clear: they see this as selling Opus-grade intelligence at a steep discount, rather than selling an expensive lightweight model. But in branding and product positioning, slapping a 4x markup on a tier known for being dirt cheap risks alienating the very developers who rely on it daily.