LLM API Pricing 2026: Full Comparison Table

Published: March 26, 2026 | Data updated: 2026-03-26 | ai-data

llm-pricingapi-costgpt-4oclaudegemini2026

LLM API Pricing Comparison 2026: Which Model Costs Least for Your Use Case?

As of March 2026, LLM API pricing spans four orders of magnitude — from $0.02/1M tokens for small open-source models to $25.00/1M tokens for top-tier flagship outputs. The cheapest model for production is Gemini 2.5 Flash-Lite at $0.10 input / $0.40 output per million tokens, while GPT-4o mini holds the crown among proprietary mid-range budget models at $0.15 input / $0.60 output. For reasoning tasks, o4-mini undercuts competitors significantly at $1.10 input / $4.40 output. If you need maximum context window with strong performance, Grok 4 offers 2M context at $3.00 / $15.00. Prices below are per 1M tokens at standard (non-batch) API rates.

TL;DR

Quick Comparison: Flagship Models

Model Provider Input $/1M Output $/1M Context Window
Claude Opus 4.6 Anthropic $5.00 $25.00 1M
GPT-5.4 OpenAI $2.50 $15.00 1.05M
GPT-5 OpenAI $1.25 $10.00 1M
Gemini 3.1 Pro Google $2.00 $12.00 1M
Grok 4 xAI $3.00 $15.00 2M
o3 OpenAI $2.00 $8.00 200K

Note: Claude Opus 4.6 is the most expensive output model in this tier. GPT-5 offers the best input cost among flagship models at $1.25/1M. Grok 4 is the only model with a 2M context window in this category.

Full LLM API Pricing Table 2026

Flagship Tier

Model Provider Input $/1M Output $/1M Context
Claude Opus 4.6 Anthropic $5.00 $25.00 1M
GPT-5.4 OpenAI $2.50 $15.00 1.05M
GPT-5 OpenAI $1.25 $10.00 1M
Gemini 3.1 Pro Google $2.00 $12.00 1M
Grok 4 xAI $3.00 $15.00 2M
o3 OpenAI $2.00 $8.00 200K

Mid-Range Tier

Model Provider Input $/1M Output $/1M Context
Claude Sonnet 4.6 Anthropic $3.00 $15.00 1M
GPT-4.1 OpenAI $2.00 $8.00 128K
GPT-4o OpenAI $2.50 $10.00 128K
o4-mini OpenAI $1.10 $4.40 200K
Gemini 2.5 Pro Google $1.25 $10.00 1M
Gemini 3 Flash Google $0.50 $3.00 1M
Mistral Large 3 Mistral $2.00 $6.00 128K
DeepSeek-R1 DeepSeek $0.55 $2.19 128K
Qwen3-Max Alibaba $0.78 $3.90 128K

Budget Tier

Model Provider Input $/1M Output $/1M Context
Claude Haiku 4.5 Anthropic $1.00 $5.00 200K
GPT-4o mini OpenAI $0.15 $0.60 128K
Gemini 2.5 Flash Google $0.30 $2.50 1M
Gemini 2.5 Flash-Lite Google $0.10 $0.40 1M
Grok 3 Mini xAI $0.30 $0.50 131K
Mistral Small 3.1 Mistral $0.10 $0.30 128K
DeepSeek-V3.2 DeepSeek $0.28 $0.42 128K
DeepSeek-V4 DeepSeek $0.30 $0.50 128K
Qwen3.5-Plus Alibaba $0.26 $1.56 128K

Open-Source Hosted (Inference Providers)

Model Provider Input $/1M Output $/1M
Llama 4 Maverick Together AI $0.27 $0.85
Llama 4 Scout Groq $0.11 $0.34
Llama 3.1 8B Groq $0.06 $0.06
Llama 3.2 3B DeepInfra $0.02 $0.02

Structured Data: Full Pricing Dataset (JSON)

{
  "dataset": "llm-api-pricing-2026",
  "updated": "2026-03-26",
  "currency": "USD",
  "unit": "per 1M tokens",
  "models": [
    {
      "id": "claude-opus-4-6",
      "name": "Claude Opus 4.6",
      "provider": "Anthropic",
      "tier": "flagship",
      "input_per_1m": 5.00,
      "output_per_1m": 25.00,
      "context_window_k": 1000,
      "batch_discount": 0.50,
      "cache_read_discount": 0.90
    },
    {
      "id": "gpt-5-4",
      "name": "GPT-5.4",
      "provider": "OpenAI",
      "tier": "flagship",
      "input_per_1m": 2.50,
      "output_per_1m": 15.00,
      "context_window_k": 1050,
      "batch_discount": 0.50,
      "cache_read_discount": 0.50
    },
    {
      "id": "gpt-5",
      "name": "GPT-5",
      "provider": "OpenAI",
      "tier": "flagship",
      "input_per_1m": 1.25,
      "output_per_1m": 10.00,
      "context_window_k": 1000,
      "batch_discount": 0.50,
      "cache_read_discount": 0.50
    },
    {
      "id": "gemini-3-1-pro",
      "name": "Gemini 3.1 Pro",
      "provider": "Google",
      "tier": "flagship",
      "input_per_1m": 2.00,
      "output_per_1m": 12.00,
      "context_window_k": 1000,
      "batch_discount": 0.50,
      "cache_read_discount": 0.90
    },
    {
      "id": "grok-4",
      "name": "Grok 4",
      "provider": "xAI",
      "tier": "flagship",
      "input_per_1m": 3.00,
      "output_per_1m": 15.00,
      "context_window_k": 2000,
      "batch_discount": null,
      "cache_read_discount": null
    },
    {
      "id": "o3",
      "name": "o3",
      "provider": "OpenAI",
      "tier": "flagship",
      "input_per_1m": 2.00,
      "output_per_1m": 8.00,
      "context_window_k": 200,
      "batch_discount": 0.50,
      "cache_read_discount": 0.50
    },
    {
      "id": "claude-sonnet-4-6",
      "name": "Claude Sonnet 4.6",
      "provider": "Anthropic",
      "tier": "mid-range",
      "input_per_1m": 3.00,
      "output_per_1m": 15.00,
      "context_window_k": 1000,
      "batch_discount": 0.50,
      "cache_read_discount": 0.90
    },
    {
      "id": "gpt-4-1",
      "name": "GPT-4.1",
      "provider": "OpenAI",
      "tier": "mid-range",
      "input_per_1m": 2.00,
      "output_per_1m": 8.00,
      "context_window_k": 128,
      "batch_discount": 0.50,
      "cache_read_discount": 0.50
    },
    {
      "id": "gpt-4o",
      "name": "GPT-4o",
      "provider": "OpenAI",
      "tier": "mid-range",
      "input_per_1m": 2.50,
      "output_per_1m": 10.00,
      "context_window_k": 128,
      "batch_discount": 0.50,
      "cache_read_discount": 0.50
    },
    {
      "id": "o4-mini",
      "name": "o4-mini",
      "provider": "OpenAI",
      "tier": "mid-range",
      "input_per_1m": 1.10,
      "output_per_1m": 4.40,
      "context_window_k": 200,
      "batch_discount": 0.50,
      "cache_read_discount": 0.50
    },
    {
      "id": "gemini-2-5-pro",
      "name": "Gemini 2.5 Pro",
      "provider": "Google",
      "tier": "mid-range",
      "input_per_1m": 1.25,
      "output_per_1m": 10.00,
      "context_window_k": 1000,
      "batch_discount": 0.50,
      "cache_read_discount": 0.90
    },
    {
      "id": "gemini-3-flash",
      "name": "Gemini 3 Flash",
      "provider": "Google",
      "tier": "mid-range",
      "input_per_1m": 0.50,
      "output_per_1m": 3.00,
      "context_window_k": 1000,
      "batch_discount": 0.50,
      "cache_read_discount": 0.90
    },
    {
      "id": "mistral-large-3",
      "name": "Mistral Large 3",
      "provider": "Mistral",
      "tier": "mid-range",
      "input_per_1m": 2.00,
      "output_per_1m": 6.00,
      "context_window_k": 128,
      "batch_discount": null,
      "cache_read_discount": null
    },
    {
      "id": "deepseek-r1",
      "name": "DeepSeek-R1",
      "provider": "DeepSeek",
      "tier": "mid-range",
      "input_per_1m": 0.55,
      "output_per_1m": 2.19,
      "context_window_k": 128,
      "batch_discount": null,
      "cache_read_discount": 0.90
    },
    {
      "id": "qwen3-max",
      "name": "Qwen3-Max",
      "provider": "Alibaba",
      "tier": "mid-range",
      "input_per_1m": 0.78,
      "output_per_1m": 3.90,
      "context_window_k": 128,
      "batch_discount": null,
      "cache_read_discount": null
    },
    {
      "id": "claude-haiku-4-5",
      "name": "Claude Haiku 4.5",
      "provider": "Anthropic",
      "tier": "budget",
      "input_per_1m": 1.00,
      "output_per_1m": 5.00,
      "context_window_k": 200,
      "batch_discount": 0.50,
      "cache_read_discount": 0.90
    },
    {
      "id": "gpt-4o-mini",
      "name": "GPT-4o mini",
      "provider": "OpenAI",
      "tier": "budget",
      "input_per_1m": 0.15,
      "output_per_1m": 0.60,
      "context_window_k": 128,
      "batch_discount": 0.50,
      "cache_read_discount": 0.50
    },
    {
      "id": "gemini-2-5-flash",
      "name": "Gemini 2.5 Flash",
      "provider": "Google",
      "tier": "budget",
      "input_per_1m": 0.30,
      "output_per_1m": 2.50,
      "context_window_k": 1000,
      "batch_discount": 0.50,
      "cache_read_discount": 0.90
    },
    {
      "id": "gemini-2-5-flash-lite",
      "name": "Gemini 2.5 Flash-Lite",
      "provider": "Google",
      "tier": "budget",
      "input_per_1m": 0.10,
      "output_per_1m": 0.40,
      "context_window_k": 1000,
      "batch_discount": 0.50,
      "cache_read_discount": 0.90
    },
    {
      "id": "grok-3-mini",
      "name": "Grok 3 Mini",
      "provider": "xAI",
      "tier": "budget",
      "input_per_1m": 0.30,
      "output_per_1m": 0.50,
      "context_window_k": 131,
      "batch_discount": null,
      "cache_read_discount": null
    },
    {
      "id": "mistral-small-3-1",
      "name": "Mistral Small 3.1",
      "provider": "Mistral",
      "tier": "budget",
      "input_per_1m": 0.10,
      "output_per_1m": 0.30,
      "context_window_k": 128,
      "batch_discount": null,
      "cache_read_discount": null
    },
    {
      "id": "deepseek-v3-2",
      "name": "DeepSeek-V3.2",
      "provider": "DeepSeek",
      "tier": "budget",
      "input_per_1m": 0.28,
      "output_per_1m": 0.42,
      "context_window_k": 128,
      "batch_discount": null,
      "cache_read_discount": 0.90
    },
    {
      "id": "deepseek-v4",
      "name": "DeepSeek-V4",
      "provider": "DeepSeek",
      "tier": "budget",
      "input_per_1m": 0.30,
      "output_per_1m": 0.50,
      "context_window_k": 128,
      "batch_discount": null,
      "cache_read_discount": 0.90
    },
    {
      "id": "qwen3-5-plus",
      "name": "Qwen3.5-Plus",
      "provider": "Alibaba",
      "tier": "budget",
      "input_per_1m": 0.26,
      "output_per_1m": 1.56,
      "context_window_k": 128,
      "batch_discount": null,
      "cache_read_discount": null
    },
    {
      "id": "llama-4-maverick",
      "name": "Llama 4 Maverick",
      "provider": "Together AI",
      "tier": "open-source-hosted",
      "input_per_1m": 0.27,
      "output_per_1m": 0.85,
      "context_window_k": null,
      "batch_discount": null,
      "cache_read_discount": null
    },
    {
      "id": "llama-4-scout",
      "name": "Llama 4 Scout",
      "provider": "Groq",
      "tier": "open-source-hosted",
      "input_per_1m": 0.11,
      "output_per_1m": 0.34,
      "context_window_k": null,
      "batch_discount": null,
      "cache_read_discount": null
    },
    {
      "id": "llama-3-1-8b",
      "name": "Llama 3.1 8B",
      "provider": "Groq",
      "tier": "open-source-hosted",
      "input_per_1m": 0.06,
      "output_per_1m": 0.06,
      "context_window_k": null,
      "batch_discount": null,
      "cache_read_discount": null
    },
    {
      "id": "llama-3-2-3b",
      "name": "Llama 3.2 3B",
      "provider": "DeepInfra",
      "tier": "open-source-hosted",
      "input_per_1m": 0.02,
      "output_per_1m": 0.02,
      "context_window_k": null,
      "batch_discount": null,
      "cache_read_discount": null
    }
  ],
  "discounts": {
    "anthropic": {
      "batch_api": "50% off standard price",
      "cache_read": "90% off standard price"
    },
    "openai": {
      "batch_api": "50% off standard price",
      "cache_read": "50% off standard price"
    },
    "google": {
      "batch_api": "50% off standard price",
      "cache_read": "90% off standard price"
    },
    "deepseek": {
      "cache_read": "90% off standard price"
    }
  }
}

Best Value Recommendations by Use Case

High-Volume Text Classification or Extraction

Use Gemini 2.5 Flash-Lite ($0.10 / $0.40) or Mistral Small 3.1 ($0.10 / $0.30). At batch scale, apply Google’s 50% discount to bring Flash-Lite to effectively $0.05 input / $0.20 output — well under $1 per 10M output tokens. For workloads where latency is not a factor, DeepInfra’s Llama 3.2 3B at $0.02 / $0.02 is the lowest available price across all hosted options.

Reasoning and Code Generation (Budget)

o4-mini ($1.10 / $4.40) is the clearest choice. It costs roughly half of o3 for input and significantly less than GPT-5 or Claude Sonnet, while retaining strong structured reasoning capabilities. Apply OpenAI’s 50% batch discount for async jobs to reach $0.55 / $2.20 effective rates.

Long-Context Document Analysis

Grok 4 (2M context, $3.00 / $15.00) is the only flagship model with a 2M token window. For 1M context needs, Claude Opus 4.6, Claude Sonnet 4.6, Gemini 2.5 Pro, and Gemini 2.5 Flash all support 1M context. Gemini 2.5 Flash at $0.30 / $2.50 is the most cost-effective 1M context option by a wide margin.

Cost-Optimized Repeated Prompts with Caching

If your application sends the same large system prompt repeatedly, cache reads can reduce costs by 90% on Anthropic and Google models. A 100K-token system prompt with Claude Sonnet 4.6 cache reads costs effectively $0.30 per 1M cached input tokens instead of $3.00. This is the single largest lever for reducing LLM API spend in production.

Open-Source with No Vendor Lock-In

Llama 4 Scout on Groq ($0.11 / $0.34) gives you a current-generation Meta model at near-budget-tier pricing. Llama 4 Maverick on Together AI ($0.27 / $0.85) offers a stronger capability tier. Both are fully compatible with the OpenAI-compatible API format used by most inference providers.

Multi-Modal or Agent Pipelines

For pipelines that require repeated short calls with tool use, GPT-4o mini ($0.15 / $0.60) remains the most battle-tested and cost-efficient option with native function calling support. DeepSeek-V3.2 ($0.28 / $0.42) offers surprisingly low output pricing for a capable model, making it attractive for high-output-ratio workloads.

API Pricing Discount Summary

Provider Batch API Discount Cache Read Discount
Anthropic 50% off 90% off
OpenAI 50% off 50% off
Google 50% off 90% off
DeepSeek N/A 90% off
xAI N/A N/A
Mistral N/A N/A
Alibaba N/A N/A

Anthropic and Google offer the most aggressive cache read discounts (90%), making them especially cost-effective for applications with repetitive system prompts or retrieval-augmented generation (RAG) workflows where retrieved chunks repeat across requests.

FAQ: LLM API Pricing 2026

What is the cheapest LLM API available in 2026?

The cheapest hosted LLM option is Llama 3.2 3B on DeepInfra at $0.02 per 1M tokens for both input and output. Among proprietary providers, Gemini 2.5 Flash-Lite and Mistral Small 3.1 are tied at $0.10/1M input. For general-purpose work with a major provider, GPT-4o mini at $0.15 / $0.60 remains one of the most cost-effective choices with reliable quality.

How do batch API discounts work and when should I use them?

Batch API (sometimes called async API) lets you submit jobs that run within a 24-hour window instead of returning results in real time. In exchange, providers offer 50% off standard pricing. This is suitable for offline data processing, document summarization, content moderation queues, and any workload without a real-time latency requirement. Anthropic, OpenAI, and Google all support batch mode. At scale, this discount alone can reduce your monthly API bill by half.

Which model has the largest context window in 2026?

Grok 4 by xAI leads with a 2M token context window at $3.00 / $15.00 per 1M tokens. Among more affordable models, Claude Opus 4.6, Claude Sonnet 4.6, Gemini 2.5 Pro, Gemini 2.5 Flash, and Gemini 2.5 Flash-Lite all support 1M token contexts. GPT models are generally capped at 128K–200K tokens.

Is DeepSeek a reliable alternative to OpenAI and Anthropic for production use?

DeepSeek models (R1, V3.2, V4) have gained significant traction in 2025-2026 as cost-effective alternatives, particularly for code and reasoning tasks. DeepSeek-R1 at $0.55 / $2.19 with 90% cache read discounts competes closely with models costing 3-4x more. However, for regulated industries or applications with data residency requirements, verify whether DeepSeek’s API infrastructure meets your compliance needs before committing to production use.

How should I choose between Claude Sonnet 4.6 and GPT-4o for my application?

Both are mid-range models with strong general capabilities. Claude Sonnet 4.6 ($3.00 / $15.00, 1M context) is more expensive but supports much longer context windows than GPT-4o ($2.50 / $10.00, 128K context). If your application involves long documents, multi-turn conversations, or large codebases, Claude Sonnet’s 1M context window provides significant architectural advantages. For shorter context workloads where cost is the primary concern, GPT-4o or GPT-4.1 are more economical.

Last updated: 2026-03-26