AI Cost Calculator
Compare API prices across GPT-5.5, Claude, Gemini, DeepSeek & more.
Estimate what it costs to build with AI — and get a model recommendation.
Prices and benchmark scores verified 2026-07-06 against public API pricing pages and independent benchmarks (LMArena, Artificial Analysis, provider model cards). Scores marked * have no published benchmark for that model/category and fall back to a rough, editorial estimate — everything else is a real, sourced number. Pricing changes often — double-check the provider's official page before committing budget.
Overall vs. price — top-left is best value
Bubble size reflects context window. Hover, tap, or focus a point for details.
19 models match your filters
Fit score for “Overall” uses a real published benchmark where one exists; scores marked * have none and fall back to a rough editorial estimate.
Llama 4 ScoutTop pick 16-expert MoE with a genuinely enormous 10M-token context window — the long-document/retrieval specialist of the Llama 4 pair. | Meta · Together | Open-hosted | $0.08 | $0.3 | 10M | 66 |
DeepSeek V4 Flash Cheap, fast open-weight with long context — roughly 35-100x cheaper than GPT-5.5 or Opus 4.8 at equivalent context. | DeepSeek | Open-hosted | $0.14 | $0.28 | 1M | 66* |
Mistral Small 4 Budget Mistral tier — newer and more capable than the older, even-cheaper Small 3.2. | Mistral | Budget | $0.15 | $0.6 | 256K | 52* |
Llama 4 Maverick 128-expert MoE, natively multimodal — the higher-quality default Llama 4 pick for customer-facing use. | Meta · Together | Open-hosted | $0.27 | $0.85 | 1M | 75 |
Codestral 25.08 Code-completion specialist — tuned for repository-level completion, not long-document tasks. | Mistral | Coding | $0.3 | $0.9 | 32K | 80* |
Mistral Large 3 80% cheaper than GPT-5.4 on input, 90% cheaper on output — Mistral's flagship value play. | Mistral | Open-hosted | $0.5 | $1.5 | 256K | 66* |
Sonar Search-augmented answer engine with live citations — not a general-purpose LLM, benchmarked differently than the models above. | Perplexity | Budget | $1 | $1 | 128K | 52* |
DeepSeek V4 Pro The best open-weight model on LMArena — within ~55 Elo points of the top closed-source models, at a fraction of the price. | DeepSeek | Open-hosted | $0.435 | $0.87 | 1M | 31 |
Grok Code Fast 1 Fast, cheap coding-focused model — optimized for iterative agentic coding speed over benchmark-topping accuracy. | xAI | Coding | $0.2 | $1.5 | 256K | 29 |
Grok 4.3 Strong on math/analytical reasoning; trails Anthropic by double digits on agentic coding benchmarks specifically. | xAI | Frontier | $1.25 | $2.5 | 1M | 53 |
Claude Haiku 4.5 Fast, cheap Claude for high-volume chat and simple tasks. | Anthropic | Budget | $1 | $5 | 200K | 81 |
Mistral Medium 3.5 Mistral's mid-flagship tier. | Mistral | Frontier | $1.5 | $7.5 | 256K | 88* |
Gemini 3.5 Flash Optimized for coding and agentic workflows — runs ~4x faster than frontier models at 25% lower cost than Gemini 3.1 Pro. | Mid-range | $1.5 | $9 | 1M | 83 | |
Claude Sonnet 5 Anthropic's new default Sonnet — now the default model on Claude Free/Pro plans, narrowing the gap to Opus 4.8 on coding and agentic work at a fraction of the price. | Anthropic | Frontier | $2 | $10 | 1M | 88* |
Gemini 3.1 Pro Google's frontier model — #3 on LMArena overall, strongest on long-context and hard reasoning in its family. | Frontier | $2 | $12 | 1M | 87 | |
Sonar Pro Perplexity's higher-quality search-augmented tier, live web search + citations included in every call. | Perplexity | Mid-range | $3 | $15 | 200K | 70* |
GPT-5.5 OpenAI's flagship — balanced coding, reasoning, and tool use. | OpenAI | Frontier | $5 | $30 | 1.05M | 89 |
Claude Opus 4.8 Anthropic's most capable model — ranks #1 overall and #1 on coding on LMArena's human-preference leaderboard. | Anthropic | Frontier | $5 | $25 | 1M | 73 |
Claude Fable 5 Anthropic's premium writing-and-reasoning flagship, priced above Opus — one source cites 95% SWE-bench Verified, the highest of any model checked this pass (unverified against a second source, flagged accordingly). | Anthropic | Frontier | $10 | $50 | 1M | 95 |
Sources: OpenAI, Anthropic, Google, DeepSeek, xAI, Mistral, Perplexity, and Together AI. Prices and context windows change frequently — treat this table as a starting point, not a live feed.
How much does AI actually cost?
There are two very different pricing models, and mixing them up is the most common source of AI sticker shock.
Subscriptions (ChatGPT Plus, Claude Pro, Gemini AI Pro, Cursor Pro, and similar) charge a flat monthly fee for hands-on use inside an app or editor. They're predictable and simple — you don't think about tokens at all — but usage is bounded by the provider's rate limits.
API pricing charges per token, quoted as dollars per 1 million tokens. This is how you pay when you're automating something or building a product that calls a model programmatically — a chatbot, a background job, an app feature. You only pay for what actually runs, but cost scales directly with volume.
Two details that catch people off guard: output tokens typically cost 3-5x more than input tokens on the same model — generating text is more expensive than reading it — and prompt caching (supported by most major providers) can cut the cost of repeated input, like a long system prompt or conversation history, by roughly 90%. Both are built into the “What it costs” tab above.