Compare LLM API pricing side by side across 31 models in 2026. GPT-5.5 costs $5/1M input, Claude 3.7 Sonnet is $3/1M, Gemini 2.0 Flash is $0.10/1M, and DeepSeek V4 Flash is the cheapest at $0.14/1M. See full output pricing, context window limits, cached input discounts, and provider comparison table.
LLM API pricing varies by up to 200x across providers in 2026. At the budget end, Gemini 2.0 Flash-Lite costs just $0.075/1M input tokens. At the premium end, Claude 3 Opus charges $15/1M input — 200 times more for the same text. Output tokens are universally more expensive than input, typically 3-6x the input price. The five major providers break down as follows: OpenAI ranges from $0.15/1M (GPT-4o Mini) to $5/1M (GPT-5.5) for input, targeting developers who want the OpenAI ecosystem. Anthropic ranges from $0.25/1M (Claude 3 Haiku) to $15/1M (Claude 3 Opus), with a focus on safety and extended thinking. Google offers the widest range from $0.0375/1M (Gemini 1.5 Flash-8B) to $1.5/1M (Gemini 3.5 Flash), all with massive context windows up to 2M tokens. DeepSeek is the price leader at $0.14/1M input for V4 Flash, offering strong reasoning at budget prices. Groq provides ultra-fast inference via Llama 4 Maverick at $0.20/1M input. Most providers offer prompt caching at 75-98% discounts on input tokens, which can dramatically reduce costs for applications with repeated system prompts. When choosing a provider, consider the total cost of ownership: API pricing, latency requirements, context window needs, and tooling support. The cheapest model per token may not be the cheapest in practice if it requires more retries, longer prompts, or additional post-processing. Use the side-by-side comparison table on this page to evaluate models across all dimensions, not just the headline per-token price.
Google Gemini 2.0 Flash-Lite at $0.075/1M input is cheapest per-token. DeepSeek V4 Flash ($0.14/1M) offers better quality. For production, GPT-4o Mini ($0.15/1M) provides the best price-to-quality ratio.
It depends on the tier. Claude 3 Haiku ($0.25/1M input) is cheaper than GPT-4o ($2.50/1M), but GPT-4o Mini ($0.15/1M) is cheaper than Claude 3.5 Haiku ($0.80/1M). Compare at the same capability level.
Prices vary up to 200x between cheapest and most expensive. Budget models cost under $0.15/1M input. Mid-range models run $1-3/1M. Premium models cost $5-15/1M. Most providers offer 50-98% discounts on cached input.
DeepSeek V4 Flash at $0.14/1M input offers the best quality-to-price ratio in 2026. GPT-4o Mini ($0.15/1M) is a close second with better tooling. For free tier usage, Gemini models offer generous no-cost quotas.
Yes. Every major provider charges separately for input (your prompt) and output (the response). Output tokens typically cost 3-6x more than input tokens. Some providers also charge differently for cached vs uncached input.
Not necessarily. Switching costs include migration effort, latency differences, and potential accuracy changes. First optimize within your current provider using caching and model routing. Only switch if the savings exceed migration costs — typically when you spend over $500/month.