Frontier LLM API Pricing: A Complete Comparison

Key Details for Each Model

Claude Fable 5

Anthropic's first generally available Mythos-class model launched on June 9, 2026, at a premium price of $10 input and $50 output per million tokens. This is exactly double the cost of Claude Opus 4.8 . The model is free to use on Pro, Max, Team, and Enterprise plans through June 22, 2026, after which metered usage credits will be billed at the API rate .

• Prompt caching: 5-minute cache writes cost $12.50/MTok, and 1-hour writes cost $20/MTok. Cache hits are $1.00/MTok .
• Positioning: Targets agentic coding, vision-heavy workloads, and long-horizon autonomous tasks .

Claude Opus 4.8

Released on May 28, 2026, Opus 4.8 keeps the same $5/$25 pricing as its predecessor Opus 4.7. A new Fast Mode is priced at $10/$50, which is three times cheaper than Opus 4.7's Fast Mode .

• Standard vs. Fast Mode: The $5/$25 standard is for regular workloads. Fast Mode at $10/$50 offers roughly 2.5x faster throughput .
• Context handling: Flat-rate pricing across the full 1M context window .

GPT-5.5

OpenAI launched GPT-5.5 on April 23, 2026, with a significant price increase—doubling the cost of GPT-5.4 to $5 per million input tokens and $30 per million output tokens . A "Pro" variant is also available at $30/$180 .

• Caching: Cached input is priced at $0.50/MTok .
• Context window: Supports up to 1,050,000 tokens with a max output of 128,000 tokens .
• Market reception: The 2x price jump was widely debated among developers .

Gemini 3.5 Flash

Google's latest Flash model went GA on May 19, 2026, at $1.50 input and $9.00 output per million tokens. This makes it approximately 40% cheaper than Gemini 3.1 Pro while delivering stronger coding and reasoning benchmarks .

• Caching: Standard tier cache hits are just $0.15/MTok, with a $1.00/MTok-per-hour storage fee .
• Regional pricing: Non-global endpoints are priced slightly higher at $1.65/$9.90 .
• Batch processing: Available at a 50% discount: $0.75 input / $4.50 output .

Grok 4.3

xAI's flagship model reached general availability on April 30, 2026, with a significant price cut. At $1.25 input and $2.50 output per million tokens, it's roughly 40% cheaper on input and 60% cheaper on output compared to its predecessor Grok 4.20 .

• Caching: Cached input is available at $0.20/MTok .
• Context window: Supports 1 million tokens, though requests exceeding 200K tokens may be billed at a higher tier .
• Reasoning: Features configurable reasoning effort levels (none, low, medium, high) .

DeepSeek V4 Pro

DeepSeek's flagship reasoning model is the undisputed cost leader. Following a permanent 75% price reduction that became official after May 31, 2026, input pricing is just $0.44/MTok and output is $0.87/MTok .

• Caching: Cache hits cost a remarkable $0.0036/MTok, making it highly economical for repetitive prompts .
• Context window: Supports a full 1M tokens with a max output of 384K tokens .

Blended Cost for a Typical Query

To help you compare total costs, we calculated a blended price per 1 million "effective tokens" using a 3:1 input-to-output ratio typical of many production workloads.

Choosing the Right Model for Your Budget

• For cost-sensitive, high-volume reasoning tasks: DeepSeek V4 Pro is the clear winner, with a blended cost more than 20x cheaper than premium alternatives.
• For a budget-friendly, general-purpose frontier model: Grok 4.3 offers competitive performance at a very low price.
• For a mid-range workhorse: Gemini 3.5 Flash balances cost and capability, performing well on coding and agent benchmarks.
• For demanding agentic pipelines and complex coding: Claude Opus 4.8 is the strong choice in the premium tier at an effective $10/MTok, while GPT-5.5 is its direct competitor at a slightly higher output price. Claude Fable 5 sits at the top as the most expensive, purpose-built for the most complex, long-horizon autonomous tasks.