Claude 3.5 Haiku Gets Cheaper: Anthropic Slashes API Costs

The Price Cut

On January 21, 2025, Anthropic announced a significant reduction in Claude 3.5 Haiku API pricing. Input tokens dropped from $1.00 to $0.80 per million tokens, while output tokens fell from $5.00 to $4.00 per million. Batch API pricing was reduced further to $0.40 input and $2.00 output per million tokens for asynchronous processing.

These changes came without any model change. Anthropic explicitly stated this is a permanent pricing reduction driven by improved inference efficiency and compute cost reductions.

What Claude 3.5 Haiku Can Do

Claude 3.5 Haiku holds an unusual position in the AI landscape: faster than most frontier models, cheaper than most mid-tier models, yet capable enough for the majority of real-world business tasks. In Anthropic's benchmarks, Haiku matches Claude 3 Opus on many coding and analysis tasks at roughly 25x lower cost per token.

Key capabilities include: 200,000 token context window, vision input support, tool use for structured function calling, and sub-second response speeds for short outputs. For most customer-facing applications, Haiku's performance is indistinguishable from more expensive models.

Competitive Context

Google had just released Gemini 2.0 Flash at $0.075 per million input tokens. OpenAI's GPT-4o Mini was at $0.15 per million. Anthropic is keeping Haiku competitive as pricing pressures intensify. Haiku defenders note it consistently requires fewer tokens to achieve the same output quality, due to superior instruction following.

Batch API and Production Use

Anthropic's batch API allows large-scale asynchronous processing at half the standard rate. At $0.40 per million input tokens in batch mode, processing 1 crore tokens costs just Rs 334 — enabling previously cost-prohibitive AI workloads like nightly data processing pipelines or large-scale content analysis.

What This Means for Indian Businesses