Anthropic Claude API Pricing History: Every Change Since Launch
Anthropic has adjusted Claude API pricing multiple times since its initial launch, and keeping track of every change is critical for teams budgeting AI costs. This article documents every confirmed model release and pricing shift from the initial Claude 1 beta through the Claude 3.5 family.
2023: The Launch Era
March 2023 — Claude 1 (API Beta)
Anthropic opened its API to early partners in March 2023. Claude 1 launched with a 9K context window at pricing that reflected limited availability:
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Context |
|---|---|---|---|
| Claude 1 | $11.02 | $32.68 | 9K |
At these rates, Claude was significantly more expensive than GPT-3.5-turbo but offered longer context and stronger instruction-following for enterprise use cases.
July 2023 — Claude 2
Claude 2 arrived with a massive 100K context window, a first for any major API provider. Pricing stayed in the same range:
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Context |
|---|---|---|---|
| Claude 2 | $11.02 | $32.68 | 100K |
The 100K context at these prices made Claude 2 the go-to choice for document analysis, legal review, and code repository understanding tasks.
November 2023 — Claude 2.1
Claude 2.1 expanded context to 200K tokens and introduced a significant price reduction:
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Context |
|---|---|---|---|
| Claude 2.1 | $8.00 | $24.00 | 200K |
This represented a roughly 27% price cut on input tokens while doubling the context window. Anthropic also introduced tool use (function calling) in beta.
2024: The Model Family Era
March 2024 — Claude 3 Family Launch
The Claude 3 launch was a watershed moment. Anthropic introduced three tiers — Haiku, Sonnet, and Opus — creating a price-performance spectrum that matched different workloads:
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Context |
|---|---|---|---|
| Claude 3 Haiku | $0.25 | $1.25 | 200K |
| Claude 3 Sonnet | $3.00 | $15.00 | 200K |
| Claude 3 Opus | $15.00 | $75.00 | 200K |
Opus set a new benchmark for reasoning quality and became the model of choice for complex analysis, while Haiku captured the fast-and-cheap routing tier.
June 2024 — Claude 3.5 Sonnet
Claude 3.5 Sonnet launched at the same price as Claude 3 Sonnet but with performance that rivaled Claude 3 Opus on most benchmarks:
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Context |
|---|---|---|---|
| Claude 3.5 Sonnet | $3.00 | $15.00 | 200K |
This was effectively a 5x price reduction for Opus-level quality, since teams could now get near-Opus performance at the Sonnet price point. Many production deployments switched from Opus to 3.5 Sonnet overnight.
October 2024 — Claude 3.5 Haiku and Sonnet Update
Anthropic released an upgraded Claude 3.5 Sonnet (new) and introduced Claude 3.5 Haiku. The Sonnet update kept the same pricing while improving coding and tool use. However, 3.5 Haiku came in at a higher price than Claude 3 Haiku:
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Context |
|---|---|---|---|
| Claude 3.5 Haiku | $0.80 | $4.00 | 200K |
| Claude 3.5 Sonnet (new) | $3.00 | $15.00 | 200K |
The 3.5 Haiku price increase (from $0.25 to $0.80 per 1M input) generated some pushback, though Anthropic argued the capability jump justified the cost. Claude 3 Haiku remained available for cost-sensitive workloads.
2025: Pricing Optimizations
Early 2025 — Claude 3.5 Sonnet Price Adjustments
Anthropic introduced prompt caching discounts and batch processing pricing, effectively reducing costs for high-volume users:
- Prompt caching: 90% reduction on cached input tokens ($0.30/1M for Sonnet cached inputs)
- Batch API: 50% reduction for asynchronous batch jobs ($1.50/1M input for Sonnet)
These discounts meant savvy teams could achieve significantly lower effective costs than the headline per-token prices.
Latest Pricing
Anthropic continues to release new models and adjust pricing. The models documented above (Claude 3 Haiku, Claude 3 Sonnet, Claude 3 Opus, Claude 3.5 Sonnet, and Claude 3.5 Haiku) have confirmed, publicly documented pricing as shown.
Key Pricing Trends
- Consistent tier pricing: Anthropic has kept Sonnet at $3/$15 and Opus at $15/$75 across the Claude 3 family. Performance improves but sticker prices stay the same, delivering more value per dollar each cycle.
- Haiku saw the only price increase: Going from $0.25 to $0.80 per 1M input between Claude 3 and 3.5 Haiku. This remains the only time Anthropic raised API prices.
- Effective cost has dropped dramatically: Prompt caching and batch pricing mean actual costs are 50-90% below headline rates for optimized workloads.
- 98% cheaper since launch: Claude 3 Haiku input pricing is 98% cheaper per token than Claude 1 was in March 2023, even before caching discounts.
- Context window stabilized at 200K: After the jump from 9K to 100K to 200K, context has held steady since Claude 3. Pricing has not varied by context usage.
How Claude Pricing Compares
In the broader AI API market, Claude's pricing is competitive across tiers:
- vs. OpenAI: OpenAI's model tiers (GPT-4o, GPT-4o mini, etc.) offer comparable pricing at each tier. Claude 3.5 Sonnet competes directly with GPT-4o on both price and performance.
- vs. Google Gemini: Gemini offers a generous free tier for development, with paid pricing comparable to the Sonnet tier. Google also offers long context windows as a competitive advantage.
- vs. Open-source (Llama, Mistral): Self-hosted models are cheaper at scale but require infrastructure investment. Claude's batch pricing makes it competitive with hosted open-source providers for many use cases.
We will continue updating this article as Anthropic releases new models and adjusts pricing. Subscribe to the iapi.buzz weekly digest to get notified of every API pricing change as it happens.