AI Model Pricing Comparison: GPT, Claude, Gemini & DeepSeek
A manually reviewed reference comparison of current LLM API pricing — input and output token cost, context scope, intended workload, and official source. Verify provider terms before choosing a production model.
Pricing is stored in a local catalog and updated only after human review. Each entry links to the provider's official pricing page.
Pricing data last checked: Aug 22, 2026 · Data sources: official provider pricing pages
| Provider | Model | Input / 1M tok | Output / 1M tok | Context window | Best use cases | Last checked | Source |
|---|---|---|---|---|---|---|---|
| OpenAI | GPT-5.6 Sol | $5 | $30 | Short-context tier | Complex professional reasoning and coding | Last checked: Aug 22, 2026 | Official |
| OpenAI | GPT-5.6 Terra | $2.5 | $15 | Short-context tier | Balanced intelligence and cost | Last checked: Aug 22, 2026 | Official |
| OpenAI | GPT-5.6 Luna | $1 | $6 | Short-context tier | Cost-sensitive, high-volume workloads | Last checked: Aug 22, 2026 | Official |
| Anthropic | Claude Opus 5 | $5 | $25 | See provider model limits | High-capability analysis and agentic work | Last checked: Aug 22, 2026 | Official |
| Anthropic | Claude Sonnet 5 | $2 | $10 | See provider model limits | General production, coding and agents | Last checked: Aug 22, 2026 | Official |
| Anthropic | Claude Haiku 4.5 | $1 | $5 | See provider model limits | Fast, lower-cost Claude workloads | Last checked: Aug 22, 2026 | Official |
Gemini 3.7 Flash | $0.75 | $3.75 | See provider model limits | Agentic workflows and multimodal reasoning | Last checked: Aug 22, 2026 | Official | |
Gemini 3.5 Flash | $1.5 | $9 | See provider model limits | Fast grounded and search-oriented workloads | Last checked: Aug 22, 2026 | Official | |
Gemini 3.5 Flash-Lite | $0.3 | $2.5 | See provider model limits | High-volume translation and simple processing | Last checked: Aug 22, 2026 | Official | |
| DeepSeek | DeepSeek V4 Flash | $0.44 | $1.32 | 1M | Low-cost general and reasoning workloads | Last checked: Aug 22, 2026 | Official |
| DeepSeek | DeepSeek V4 Pro | $1.32 | $3.96 | 1M | Higher-capability DeepSeek workloads | Last checked: Aug 22, 2026 | Official |
Disclaimer: Prices are manually reviewed reference values in USD per one million tokens and are not real-time. Always verify current pricing on each provider's official page before making a purchasing decision.
The cheapest model is not always the best choice
A low per-token price looks attractive on a spreadsheet, but production AI costs are shaped by more than the rate card. Weigh these three tradeoffs before locking in a model for your application.
Accuracy vs. cost
Cheaper models often need more retries, longer prompts, or extra verification steps to reach the same output quality. A model with a higher token price but fewer failed generations can end up cheaper per successful task — and it protects the user experience.
Latency vs. cost
Smaller, cheaper models are usually faster, which matters for real-time chat or agent loops. Larger flagship models can be slower even though they reason better — so the "best" model depends on whether your product needs instant responses or can tolerate a few extra seconds for higher-quality output.
Production requirements
Rate limits, context window size, function-calling support, uptime, and data-handling policies all differ by provider and tier. A model that's cheap in isolated testing can become the expensive choice once you factor in outages, throttling, or the engineering time needed to work around its limitations at scale.
Rule of thumb: use a fast, cheap model for high-volume, low-risk tasks (classification, extraction, simple chat), and reserve a higher-accuracy model for the steps where a mistake is costly — then measure real cost per successful outcome, not just cost per token.
See what this pricing means for your workload
Plug your token counts into the calculator to see real cost per request and per month for any model above.