LLM API Pricing in June 2026: Input vs Output Token Cost Breakdown

LLM API pricing follows a consistent structure: you pay separately for input tokens (your prompt) and output tokens (the model’s response). Output tokens are almost always more expensive than input tokens because generating each output token requires one full forward pass through the model, while input tokens are processed in parallel. Understanding the ratio matters as much as the headline price.

Use our LLM cost calculator to estimate costs for your specific prompt and response lengths using the prices in this table.

Pricing table — June 2026

All prices are in USD per 1 million tokens ($/1M). Prices are from each provider’s public pricing page as of June 2026 and are subject to change. Standard (non-batch, non-cached) API pricing is shown.

Prices verified from provider pricing pages in June 2026. Always check the provider’s current pricing page before committing to a production budget.

Output/input ratio analysis

The output/input cost ratio reveals the pricing philosophy of each provider. Four patterns are visible in the data:

• 4× ratio (OpenAI, Gemini Flash, Cohere). The most common ratio. Reflects the computational asymmetry of autoregressive generation vs. parallel prefill.
• 5× ratio (Anthropic Claude). Anthropic charges a higher output premium, consistent with their longer-average-output benchmarks and extended thinking capabilities.
• 8× ratio (Gemini 2.5 Pro).The highest ratio in our table — Gemini 2.5 Pro’s output premium reflects its extended reasoning and long-context generation capabilities.
• 1× ratio (Meta Llama via Together.ai). Open-weight models hosted on inference providers often charge the same rate for input and output, treating tokens as a commodity compute unit rather than differentiating by generation direction.

For workloads where output is long relative to input (e.g., document generation, code synthesis, summarization), the output token rate dominates total cost. For RAG pipelines with large context windows and short responses, the input rate dominates.

Cost per 10,000 tokens

10,000 tokens is approximately 7,500 words of English prose — a short article, a medium code file, or a multi-turn conversation. At 100 tokens output per 1,000 tokens input (a typical RAG-style ratio):

Use the LLM cost calculator to model your actual prompt/output ratio and compare total monthly cost across providers.

Key observations

• Cheapest per token (input): Gemini 1.5 Flash at $0.075/1M — less than a hundredth of a cent per 1,000 tokens.
• Most expensive per token (output): o3 at $40/1M output — 533× more expensive than Gemini 1.5 Flash output at $0.30/1M.
• Best cost parity (input = output): Meta Llama 3.3 70B via Together.ai at $0.18/1M both ways. Suitable for symmetric workloads.
• Best cost for long-context RAG: Gemini 1.5 Flash — both the lowest input price and 1M token context window.
• Prompt caching discounts (not shown) can reduce effective input costs by 50–90% for repeated system prompts. Anthropic, OpenAI (Batch API), and Google all offer caching or batch discounts that materially change the economics for high-volume production use.

Limitations and caveats

• Prices change frequently. LLM API pricing has decreased ~10× per year for comparable capability since 2023. The values in this table reflect public pricing as of June 2026 and may already be outdated at time of reading.
• Quality is not equal. A lower price per token does not mean lower total cost if the model requires longer prompts for equivalent task performance, more retries, or post-processing.
• Throughput and latency vary.Gemini 1.5 Flash’s low price comes with shared quota limits; provisioned throughput on GPT-4o costs more per token but guarantees capacity.
• Fine-tuning, embedding, and image tokens are excluded. This table covers text generation only.

Sources

OpenAI pricing page (openai.com/api/pricing); Anthropic API pricing (anthropic.com/pricing); Google AI Studio and Vertex AI pricing (ai.google.dev/pricing); Together.ai pricing (together.ai/pricing); Mistral AI pricing (mistral.ai/technology); Cohere pricing (cohere.com/pricing). All prices verified June 2026.