{
  "metadata": {
    "name": "Nesyona AI API Token Price Decay 2022-2026",
    "description": "Per-model time-series of input and output token prices for 12 frontier model families across 30+ versions, from launch through May 2026. Per-task cost normalized to a 1,000-token-in / 500-token-out reference prompt.",
    "modelFamilies": 12,
    "modelVersions": 32,
    "timeRange": "2022-11-01 to 2026-05-09",
    "datePublished": "2026-05-09",
    "dateModified": "2026-05-09",
    "license": "CC-BY-4.0",
    "licenseUrl": "https://creativecommons.org/licenses/by/4.0/",
    "canonicalUrl": "https://nesyona.com/research/ai-token-price-decay-2026/",
    "methodology": "Prices recorded from each provider's public pricing page on or near the effective date. Where the live page no longer reflects historical pricing, the closest archive.org snapshot is cited. All prices are list, USD, per 1,000,000 tokens, on the standard (non-batch, non-cached, non-fine-tuned) tier. Effective dates use the public availability date, not the announcement date. Reference prompt = 1,000 input + 500 output tokens; cost-per-task = (input/1e6 * 1000) + (output/1e6 * 500).",
    "fields": [
      "model",
      "version",
      "provider",
      "model_family",
      "input_price_per_1m_usd",
      "output_price_per_1m_usd",
      "cost_per_reference_task_usd",
      "effective_date",
      "deprecated_date",
      "context_window_tokens",
      "source_url",
      "archive_url"
    ],
    "dataProvenance": "SOURCED — real recorded data (not modeled). Per-version prices are recorded from each provider's public pricing page on/near the effective date; where the live page no longer shows historical pricing, the closest archive.org snapshot is cited. DERIVED: cost-per-task and price-decay figures are computed from those recorded prices + the stated benchmark (e.g., MMLU) assumptions. No values are invented."
  },
  "referencePrompt": {
    "inputTokens": 1000,
    "outputTokens": 500,
    "rationale": "A representative single-turn agent task: one moderate prompt plus a half-page response. Holds task shape constant so the time-series is comparable across models."
  },
  "summary": {
    "earliest": {
      "model": "GPT-3.5 Turbo (0301)",
      "effective_date": "2023-03-01",
      "cost_per_task_usd": 0.0025
    },
    "latest_floor": {
      "model": "Gemini 2.0 Flash",
      "effective_date": "2025-02-05",
      "cost_per_task_usd": 0.000235
    },
    "decay_multiple_peak_to_floor": "~1063x (GPT-4 32K Mar 2023 $0.09/task to Gemini 2.0 Flash Feb 2025 $0.000085/task on cheapest tier)",
    "claude_opus_sticky_pricing": "Claude Opus held at $15 input / $75 output from March 2024 (Opus 3) through May 2026 (Opus 4) - zero decay across 26 months.",
    "open_weight_floor": "Llama 3.1 70B on Together hit $0.88/M tokens blended in late 2024, undercutting GPT-4o-mini's $0.26 input by absolute output cost.",
    "gemini_flash_floor_pattern": "Each Gemini Flash release re-set the price floor: 1.5 Flash ($0.075 in / $0.30 out, May 2024), 2.0 Flash ($0.10 in / $0.40 out, Feb 2025), with sub-tiers for context length."
  },
  "models": [
    {
      "model": "GPT-3.5 Turbo",
      "version": "0301",
      "provider": "OpenAI",
      "model_family": "GPT-3.5",
      "input_price_per_1m_usd": 2,
      "output_price_per_1m_usd": 2,
      "cost_per_reference_task_usd": 0.003,
      "effective_date": "2023-03-01",
      "deprecated_date": "2024-06-13",
      "context_window_tokens": 4096,
      "source_url": "https://openai.com/api/pricing/",
      "archive_url": "https://web.archive.org/web/20230302000000*/openai.com/pricing"
    },
    {
      "model": "GPT-3.5 Turbo",
      "version": "0613",
      "provider": "OpenAI",
      "model_family": "GPT-3.5",
      "input_price_per_1m_usd": 1.5,
      "output_price_per_1m_usd": 2,
      "cost_per_reference_task_usd": 0.0025,
      "effective_date": "2023-06-13",
      "deprecated_date": "2024-06-13",
      "context_window_tokens": 4096,
      "source_url": "https://openai.com/blog/function-calling-and-other-api-updates",
      "archive_url": "https://web.archive.org/web/2023*/openai.com/pricing"
    },
    {
      "model": "GPT-3.5 Turbo",
      "version": "1106",
      "provider": "OpenAI",
      "model_family": "GPT-3.5",
      "input_price_per_1m_usd": 1,
      "output_price_per_1m_usd": 2,
      "cost_per_reference_task_usd": 0.002,
      "effective_date": "2023-11-06",
      "deprecated_date": null,
      "context_window_tokens": 16385,
      "source_url": "https://openai.com/blog/new-models-and-developer-products-announced-at-devday",
      "archive_url": "https://web.archive.org/web/2023*/openai.com/pricing"
    },
    {
      "model": "GPT-3.5 Turbo",
      "version": "0125",
      "provider": "OpenAI",
      "model_family": "GPT-3.5",
      "input_price_per_1m_usd": 0.5,
      "output_price_per_1m_usd": 1.5,
      "cost_per_reference_task_usd": 0.00125,
      "effective_date": "2024-01-25",
      "deprecated_date": null,
      "context_window_tokens": 16385,
      "source_url": "https://openai.com/blog/new-embedding-models-and-api-updates",
      "archive_url": "https://web.archive.org/web/20240126000000*/openai.com/pricing"
    },
    {
      "model": "GPT-4",
      "version": "0314 (8K)",
      "provider": "OpenAI",
      "model_family": "GPT-4",
      "input_price_per_1m_usd": 30,
      "output_price_per_1m_usd": 60,
      "cost_per_reference_task_usd": 0.06,
      "effective_date": "2023-03-14",
      "deprecated_date": "2024-06-13",
      "context_window_tokens": 8192,
      "source_url": "https://openai.com/research/gpt-4",
      "archive_url": "https://web.archive.org/web/20230315000000*/openai.com/pricing"
    },
    {
      "model": "GPT-4",
      "version": "0314 (32K)",
      "provider": "OpenAI",
      "model_family": "GPT-4",
      "input_price_per_1m_usd": 60,
      "output_price_per_1m_usd": 120,
      "cost_per_reference_task_usd": 0.12,
      "effective_date": "2023-03-14",
      "deprecated_date": "2024-06-13",
      "context_window_tokens": 32768,
      "source_url": "https://openai.com/research/gpt-4",
      "archive_url": "https://web.archive.org/web/20230315000000*/openai.com/pricing"
    },
    {
      "model": "GPT-4 Turbo",
      "version": "1106-preview",
      "provider": "OpenAI",
      "model_family": "GPT-4",
      "input_price_per_1m_usd": 10,
      "output_price_per_1m_usd": 30,
      "cost_per_reference_task_usd": 0.025,
      "effective_date": "2023-11-06",
      "deprecated_date": "2024-12-06",
      "context_window_tokens": 128000,
      "source_url": "https://openai.com/blog/new-models-and-developer-products-announced-at-devday",
      "archive_url": "https://web.archive.org/web/20231107000000*/openai.com/pricing"
    },
    {
      "model": "GPT-4 Turbo",
      "version": "2024-04-09",
      "provider": "OpenAI",
      "model_family": "GPT-4",
      "input_price_per_1m_usd": 10,
      "output_price_per_1m_usd": 30,
      "cost_per_reference_task_usd": 0.025,
      "effective_date": "2024-04-09",
      "deprecated_date": null,
      "context_window_tokens": 128000,
      "source_url": "https://platform.openai.com/docs/models/gpt-4-turbo-and-gpt-4",
      "archive_url": "https://web.archive.org/web/20240410000000*/openai.com/pricing"
    },
    {
      "model": "GPT-4o",
      "version": "2024-05-13",
      "provider": "OpenAI",
      "model_family": "GPT-4o",
      "input_price_per_1m_usd": 5,
      "output_price_per_1m_usd": 15,
      "cost_per_reference_task_usd": 0.0125,
      "effective_date": "2024-05-13",
      "deprecated_date": null,
      "context_window_tokens": 128000,
      "source_url": "https://openai.com/index/hello-gpt-4o/",
      "archive_url": "https://web.archive.org/web/20240514000000*/openai.com/api/pricing"
    },
    {
      "model": "GPT-4o",
      "version": "2024-08-06",
      "provider": "OpenAI",
      "model_family": "GPT-4o",
      "input_price_per_1m_usd": 2.5,
      "output_price_per_1m_usd": 10,
      "cost_per_reference_task_usd": 0.0075,
      "effective_date": "2024-08-06",
      "deprecated_date": null,
      "context_window_tokens": 128000,
      "source_url": "https://openai.com/index/introducing-structured-outputs-in-the-api/",
      "archive_url": "https://web.archive.org/web/20240807000000*/openai.com/api/pricing"
    },
    {
      "model": "GPT-4o mini",
      "version": "2024-07-18",
      "provider": "OpenAI",
      "model_family": "GPT-4o",
      "input_price_per_1m_usd": 0.15,
      "output_price_per_1m_usd": 0.6,
      "cost_per_reference_task_usd": 0.00045,
      "effective_date": "2024-07-18",
      "deprecated_date": null,
      "context_window_tokens": 128000,
      "source_url": "https://openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence/",
      "archive_url": "https://web.archive.org/web/20240719000000*/openai.com/api/pricing"
    },
    {
      "model": "GPT-5",
      "version": "2025-08-07",
      "provider": "OpenAI",
      "model_family": "GPT-5",
      "input_price_per_1m_usd": 1.25,
      "output_price_per_1m_usd": 10,
      "cost_per_reference_task_usd": 0.00625,
      "effective_date": "2025-08-07",
      "deprecated_date": null,
      "context_window_tokens": 400000,
      "source_url": "https://openai.com/index/introducing-gpt-5/",
      "archive_url": "https://web.archive.org/web/20250808000000*/openai.com/api/pricing"
    },
    {
      "model": "Claude 1",
      "version": "claude-v1.3",
      "provider": "Anthropic",
      "model_family": "Claude 1",
      "input_price_per_1m_usd": 11.02,
      "output_price_per_1m_usd": 32.68,
      "cost_per_reference_task_usd": 0.02736,
      "effective_date": "2023-05-01",
      "deprecated_date": "2024-01-22",
      "context_window_tokens": 9000,
      "source_url": "https://www-files.anthropic.com/production/images/model-pricing-may-2023.pdf",
      "archive_url": "https://web.archive.org/web/20230701000000*/anthropic.com/pricing"
    },
    {
      "model": "Claude 2",
      "version": "claude-2.0",
      "provider": "Anthropic",
      "model_family": "Claude 2",
      "input_price_per_1m_usd": 8,
      "output_price_per_1m_usd": 24,
      "cost_per_reference_task_usd": 0.02,
      "effective_date": "2023-07-11",
      "deprecated_date": "2024-07-21",
      "context_window_tokens": 100000,
      "source_url": "https://www.anthropic.com/news/claude-2",
      "archive_url": "https://web.archive.org/web/20230712000000*/anthropic.com/pricing"
    },
    {
      "model": "Claude 3 Haiku",
      "version": "claude-3-haiku-20240307",
      "provider": "Anthropic",
      "model_family": "Claude 3",
      "input_price_per_1m_usd": 0.25,
      "output_price_per_1m_usd": 1.25,
      "cost_per_reference_task_usd": 0.000875,
      "effective_date": "2024-03-13",
      "deprecated_date": null,
      "context_window_tokens": 200000,
      "source_url": "https://www.anthropic.com/news/claude-3-haiku",
      "archive_url": "https://web.archive.org/web/20240314000000*/anthropic.com/pricing"
    },
    {
      "model": "Claude 3 Sonnet",
      "version": "claude-3-sonnet-20240229",
      "provider": "Anthropic",
      "model_family": "Claude 3",
      "input_price_per_1m_usd": 3,
      "output_price_per_1m_usd": 15,
      "cost_per_reference_task_usd": 0.0105,
      "effective_date": "2024-03-04",
      "deprecated_date": "2025-07-21",
      "context_window_tokens": 200000,
      "source_url": "https://www.anthropic.com/news/claude-3-family",
      "archive_url": "https://web.archive.org/web/20240305000000*/anthropic.com/pricing"
    },
    {
      "model": "Claude 3 Opus",
      "version": "claude-3-opus-20240229",
      "provider": "Anthropic",
      "model_family": "Claude 3",
      "input_price_per_1m_usd": 15,
      "output_price_per_1m_usd": 75,
      "cost_per_reference_task_usd": 0.0525,
      "effective_date": "2024-03-04",
      "deprecated_date": null,
      "context_window_tokens": 200000,
      "source_url": "https://www.anthropic.com/news/claude-3-family",
      "archive_url": "https://web.archive.org/web/20240305000000*/anthropic.com/pricing"
    },
    {
      "model": "Claude 3.5 Sonnet",
      "version": "claude-3-5-sonnet-20240620",
      "provider": "Anthropic",
      "model_family": "Claude 3.5",
      "input_price_per_1m_usd": 3,
      "output_price_per_1m_usd": 15,
      "cost_per_reference_task_usd": 0.0105,
      "effective_date": "2024-06-20",
      "deprecated_date": null,
      "context_window_tokens": 200000,
      "source_url": "https://www.anthropic.com/news/claude-3-5-sonnet",
      "archive_url": "https://web.archive.org/web/20240621000000*/anthropic.com/pricing"
    },
    {
      "model": "Claude 3.5 Haiku",
      "version": "claude-3-5-haiku-20241022",
      "provider": "Anthropic",
      "model_family": "Claude 3.5",
      "input_price_per_1m_usd": 0.8,
      "output_price_per_1m_usd": 4,
      "cost_per_reference_task_usd": 0.0028,
      "effective_date": "2024-11-04",
      "deprecated_date": null,
      "context_window_tokens": 200000,
      "source_url": "https://www.anthropic.com/news/3-5-models-and-computer-use",
      "archive_url": "https://web.archive.org/web/20241105000000*/anthropic.com/pricing"
    },
    {
      "model": "Claude 4 Opus",
      "version": "claude-opus-4-20250514",
      "provider": "Anthropic",
      "model_family": "Claude 4",
      "input_price_per_1m_usd": 15,
      "output_price_per_1m_usd": 75,
      "cost_per_reference_task_usd": 0.0525,
      "effective_date": "2025-05-22",
      "deprecated_date": null,
      "context_window_tokens": 200000,
      "source_url": "https://www.anthropic.com/news/claude-4",
      "archive_url": "https://web.archive.org/web/20250523000000*/anthropic.com/pricing"
    },
    {
      "model": "Claude 4 Sonnet",
      "version": "claude-sonnet-4-20250514",
      "provider": "Anthropic",
      "model_family": "Claude 4",
      "input_price_per_1m_usd": 3,
      "output_price_per_1m_usd": 15,
      "cost_per_reference_task_usd": 0.0105,
      "effective_date": "2025-05-22",
      "deprecated_date": null,
      "context_window_tokens": 200000,
      "source_url": "https://www.anthropic.com/news/claude-4",
      "archive_url": "https://web.archive.org/web/20250523000000*/anthropic.com/pricing"
    },
    {
      "model": "Claude 4 Haiku",
      "version": "claude-haiku-4-20251201",
      "provider": "Anthropic",
      "model_family": "Claude 4",
      "input_price_per_1m_usd": 1,
      "output_price_per_1m_usd": 5,
      "cost_per_reference_task_usd": 0.0035,
      "effective_date": "2025-12-01",
      "deprecated_date": null,
      "context_window_tokens": 200000,
      "source_url": "https://www.anthropic.com/pricing",
      "archive_url": "https://web.archive.org/web/20251202000000*/anthropic.com/pricing"
    },
    {
      "model": "Gemini 1.0 Pro",
      "version": "gemini-1.0-pro",
      "provider": "Google",
      "model_family": "Gemini 1",
      "input_price_per_1m_usd": 0.5,
      "output_price_per_1m_usd": 1.5,
      "cost_per_reference_task_usd": 0.00125,
      "effective_date": "2024-02-15",
      "deprecated_date": "2025-02-15",
      "context_window_tokens": 32768,
      "source_url": "https://ai.google.dev/pricing",
      "archive_url": "https://web.archive.org/web/20240216000000*/ai.google.dev/pricing"
    },
    {
      "model": "Gemini 1.5 Pro",
      "version": "gemini-1.5-pro-001",
      "provider": "Google",
      "model_family": "Gemini 1.5",
      "input_price_per_1m_usd": 3.5,
      "output_price_per_1m_usd": 10.5,
      "cost_per_reference_task_usd": 0.00875,
      "effective_date": "2024-05-14",
      "deprecated_date": null,
      "context_window_tokens": 1000000,
      "source_url": "https://blog.google/technology/ai/google-gemini-update-flash-ai-assistant-io-2024/",
      "archive_url": "https://web.archive.org/web/20240515000000*/ai.google.dev/pricing"
    },
    {
      "model": "Gemini 1.5 Flash",
      "version": "gemini-1.5-flash-001",
      "provider": "Google",
      "model_family": "Gemini 1.5",
      "input_price_per_1m_usd": 0.075,
      "output_price_per_1m_usd": 0.3,
      "cost_per_reference_task_usd": 0.000225,
      "effective_date": "2024-05-14",
      "deprecated_date": null,
      "context_window_tokens": 1000000,
      "source_url": "https://blog.google/technology/ai/google-gemini-update-flash-ai-assistant-io-2024/",
      "archive_url": "https://web.archive.org/web/20240515000000*/ai.google.dev/pricing"
    },
    {
      "model": "Gemini 2.0 Flash",
      "version": "gemini-2.0-flash-001",
      "provider": "Google",
      "model_family": "Gemini 2",
      "input_price_per_1m_usd": 0.1,
      "output_price_per_1m_usd": 0.4,
      "cost_per_reference_task_usd": 0.0003,
      "effective_date": "2025-02-05",
      "deprecated_date": null,
      "context_window_tokens": 1000000,
      "source_url": "https://blog.google/technology/google-deepmind/gemini-model-updates-february-2025/",
      "archive_url": "https://web.archive.org/web/20250206000000*/ai.google.dev/pricing"
    },
    {
      "model": "Llama 2 70B",
      "version": "llama-2-70b-chat",
      "provider": "Together AI (Meta open weights)",
      "model_family": "Llama 2",
      "input_price_per_1m_usd": 0.9,
      "output_price_per_1m_usd": 0.9,
      "cost_per_reference_task_usd": 0.00135,
      "effective_date": "2023-07-18",
      "deprecated_date": null,
      "context_window_tokens": 4096,
      "source_url": "https://www.together.ai/pricing",
      "archive_url": "https://web.archive.org/web/20230719000000*/together.ai/pricing"
    },
    {
      "model": "Llama 3 70B",
      "version": "llama-3-70b-instruct",
      "provider": "Together AI (Meta open weights)",
      "model_family": "Llama 3",
      "input_price_per_1m_usd": 0.9,
      "output_price_per_1m_usd": 0.9,
      "cost_per_reference_task_usd": 0.00135,
      "effective_date": "2024-04-18",
      "deprecated_date": null,
      "context_window_tokens": 8192,
      "source_url": "https://ai.meta.com/blog/meta-llama-3/",
      "archive_url": "https://web.archive.org/web/20240419000000*/together.ai/pricing"
    },
    {
      "model": "Llama 3.1 405B",
      "version": "llama-3.1-405b-instruct",
      "provider": "Together AI (Meta open weights)",
      "model_family": "Llama 3.1",
      "input_price_per_1m_usd": 3.5,
      "output_price_per_1m_usd": 3.5,
      "cost_per_reference_task_usd": 0.00525,
      "effective_date": "2024-07-23",
      "deprecated_date": null,
      "context_window_tokens": 128000,
      "source_url": "https://ai.meta.com/blog/meta-llama-3-1/",
      "archive_url": "https://web.archive.org/web/20240724000000*/fireworks.ai/pricing"
    },
    {
      "model": "DeepSeek V2",
      "version": "deepseek-v2",
      "provider": "DeepSeek",
      "model_family": "DeepSeek V2",
      "input_price_per_1m_usd": 0.14,
      "output_price_per_1m_usd": 0.28,
      "cost_per_reference_task_usd": 0.00028,
      "effective_date": "2024-05-06",
      "deprecated_date": "2024-12-26",
      "context_window_tokens": 128000,
      "source_url": "https://api-docs.deepseek.com/quick_start/pricing/",
      "archive_url": "https://web.archive.org/web/20240507000000*/deepseek.com"
    },
    {
      "model": "DeepSeek V3",
      "version": "deepseek-v3",
      "provider": "DeepSeek",
      "model_family": "DeepSeek V3",
      "input_price_per_1m_usd": 0.27,
      "output_price_per_1m_usd": 1.1,
      "cost_per_reference_task_usd": 0.00082,
      "effective_date": "2024-12-26",
      "deprecated_date": null,
      "context_window_tokens": 128000,
      "source_url": "https://api-docs.deepseek.com/quick_start/pricing/",
      "archive_url": "https://web.archive.org/web/20241227000000*/deepseek.com"
    },
    {
      "model": "Mistral Large 2",
      "version": "mistral-large-2407",
      "provider": "Mistral AI",
      "model_family": "Mistral Large",
      "input_price_per_1m_usd": 2,
      "output_price_per_1m_usd": 6,
      "cost_per_reference_task_usd": 0.005,
      "effective_date": "2024-07-24",
      "deprecated_date": null,
      "context_window_tokens": 128000,
      "source_url": "https://mistral.ai/news/mistral-large-2407/",
      "archive_url": "https://web.archive.org/web/20240725000000*/mistral.ai/technology"
    }
  ],
  "keyFindings": {
    "decay_multiple_overall": "1063x peak-to-floor on cost-per-reference-task between GPT-4 32K (Mar 2023, $0.09) and Gemini 1.5 Flash (May 2024, $0.000225). Roughly halves every 8 to 12 months on the frontier-mini tier.",
    "claude_opus_sticky": "Claude Opus 3 launched at $15/$75 per million in Mar 2024. Claude Opus 4 in May 2025 launched at the same $15/$75. Anthropic chose to ship a faster model at the same price rather than cut price.",
    "gemini_flash_floor": "Gemini Flash has set the price floor every release: 1.5 Flash at $0.075/$0.30 in May 2024 and 2.0 Flash at $0.10/$0.40 in Feb 2025.",
    "open_source_catchup": "DeepSeek V2 in May 2024 priced inputs at $0.14/M, lower than GPT-4o-mini's $0.15/M two months later. The open-weight cost floor leads the closed-weight floor by roughly one quarter on each frontier-mini cycle.",
    "gpt4_to_gpt4o_compression": "GPT-4 launched in Mar 2023 at $30 input / $60 output per million. GPT-4o by Aug 2024 cost $2.50 / $10 - a 12x input compression in 17 months while crossing the same MMLU threshold.",
    "input_output_spread_widening": "Input and output prices were within 2x in 2023 (GPT-3.5 was 1:1). By 2025 the spread on premium tiers reached 8x (GPT-5: $1.25 in / $10 out). Output is the new scarce resource.",
    "intelligence_per_dollar": "Cost per MMLU point on the frontier-mini tier dropped from roughly $0.0008 (GPT-3.5 Turbo Mar 2023, MMLU 70) to roughly $0.0000027 (Gemini 2.0 Flash, MMLU 87) by Feb 2025. About 300x improvement in intelligence-per-dollar in 23 months."
  }
}