LLM token cost calculatorCalculadora de costos de tokens LLM

Applied Agentic AI for Organizational Transformation · Top 10 frontier models · November 2026Top 10 modelos frontera · Noviembre 2026

Workload presets:Perfiles de carga: i

Or size it from a real interaction — both sliders follow your text O calcula desde una interacción real — ambos deslizadores siguen tu texto i

Popular prompts:Prompts populares:

Paste a prompt or pick a sample above.Pega un prompt o elige un ejemplo arriba.

Paste an answer or pick a sample above.Pega una respuesta o elige un ejemplo arriba.

CheapestMás económico

—

—

Most expensiveMás costoso

—

—

Price spreadDiferencial de precio

—

cheapest vs. priciest for this workloadmás económico vs. más costoso para esta carga

Best value, adjustedMejor valor, ajustado

—

—

Largest contextMayor contexto

—

—

FastestMás rápido

—

—

Model comparisonComparación de modelos

Rank by:Ordenar por: i
ModelModelo ContextContexto i Input $/1MEntrada $/1M i Cached $/1MCaché $/1M i Output $/1MSalida $/1M i Speed (t/s)Velocidad (t/s) i OverheadOverhead i Cost / monthCosto / mes i
Cheapest per token is not cheapest per task. Reasoning models spend output tokens "thinking," and some spend far more than others to finish the same work: Claude Opus 5.5 generated ~260M output tokens on the Artificial Analysis Intelligence Index v4.3 versus GPT-5.6 Sol's ~90M reference on the same evaluation (2.9× thinking overhead) [49] [54]. The efficiency-adjusted ranking multiplies each model's output tokens by its thinking overhead — watch how the gap between "cheap" and "expensive" models narrows or flips.
Lo más económico por token no es lo más económico por tarea. Los modelos de razonamiento consumen tokens de salida "pensando", y algunos gastan mucho más que otros para finalizar el mismo trabajo: Claude Opus 5.5 generó ~260M de tokens de salida en el Índice de Inteligencia v4.3 de Artificial Analysis frente a los ~90M de referencia de GPT-5.6 Sol en la misma evaluación (2.9× de overhead de razonamiento) [49] [54]. La clasificación ajustada por eficiencia multiplica los tokens de salida de cada modelo por su overhead de pensamiento — observa cómo la brecha entre modelos "económicos" y "costosos" se reduce o se invierte.
Why Meta is now included: Meta has published per-token API pricing for its Muse Spark family since Muse Spark 1.1's July 2026 launch — $1.25 / $4.25 per million input/output tokens, $0.15 cached — and kept those rates for Muse Spark 1.3 in October 2026 [16] [57] [58]. Muse Spark 1.3 scores 48 on the Artificial Analysis Intelligence Index v4.3 [57], ahead of Kimi K3 (44) [28], and costs less at every rate with faster output, so it takes Kimi K3's place. Selection rule: a mix of the highest-capability and best-value models with published per-token pricing, maximum two per company.
Por qué ahora se incluye Meta: Meta ha publicado precios de API por token para su familia Muse Spark desde el lanzamiento de Muse Spark 1.1 en julio de 2026 — $1.25 / $4.25 por millón de tokens de entrada/salida, $0.15 en caché — y mantuvo esas tarifas para Muse Spark 1.3 en octubre de 2026 [16] [57] [58]. Muse Spark 1.3 obtiene 48 en el Índice de Inteligencia v4.3 de Artificial Analysis [57], por encima de Kimi K3 (44) [28], y cuesta menos en todas las tarifas con salida más rápida, por lo que ocupa el lugar de Kimi K3. Regla de selección: una combinación de modelos de máxima capacidad y mejor relación calidad-precio con precios publicados por token, máximo dos por empresa.
Roster update, October 19, 2026: GPT-6 Astra replaces GPT-5.5 Pro. It is OpenAI's current flagship at $10 / $50 per million tokens with a $1 cache-read rate — a third of GPT-5.5 Pro's $30 / $180, which had no cache discount — and scores 53 on the Artificial Analysis Intelligence Index (#7) [59] [60] [61]. Gemini 4 Argon replaces Gemini 3.1 Pro: Google's new frontier model scores 53 (#8) at an introductory $2 / $10 per million ($0.10 cached) through November 30, 2026; from December 1 the list price doubles to $4 / $20 ($0.20 cached) [62] [63]. Prompts above 272K input tokens on GPT-6 Astra bill at 2× input and 1.5× output [59]. Artificial Analysis has not yet published an output speed for Argon; the figure shown is OpenRouter's P50 throughput on Google's own endpoints [64].
Actualización de la lista, 19 de octubre de 2026: GPT-6 Astra sustituye a GPT-5.5 Pro. Es el modelo insignia actual de OpenAI a $10 / $50 por millón de tokens con lectura de caché a $1 — una tercera parte de los $30 / $180 de GPT-5.5 Pro, que no tenía descuento de caché — y obtiene 53 en el Índice de Inteligencia de Artificial Analysis (#7) [59] [60] [61]. Gemini 4 Argon sustituye a Gemini 3.1 Pro: el nuevo modelo frontera de Google obtiene 53 (#8) a un precio introductorio de $2 / $10 por millón ($0.10 en caché) hasta el 30 de noviembre de 2026; desde el 1 de diciembre el precio de lista sube al doble, $4 / $20 ($0.20 en caché) [62] [63]. En GPT-6 Astra, los prompts con más de 272K tokens de entrada se cobran a 2× la entrada y 1.5× la salida [59]. Artificial Analysis aún no publica velocidad de salida para Argon; la cifra mostrada es el rendimiento P50 de OpenRouter en los endpoints de Google [64].

ReferencesReferencias

Los títulos de las fuentes se citan en su idioma original.

  1. Anthropic — Claude platform pricing
  2. Anthropic — prompt caching (cache reads = 0.1× input price)
  3. OpenRouter — Claude Fable 5 (pricing, 1M context)
  4. OpenAI — Introducing GPT-5.5 (GPT-5.5 Pro pricing)
  5. OpenAI — API pricing
  6. OpenRouter — GPT-5.5 (1.05M context)
  7. Google — Gemini API pricing
  8. Simon Willison — Gemini 3.5 Flash pricing analysis
  9. xAI — model documentation
  10. xAI — Grok 4.20 model card (1M context, $0.20/1M cached input)
  11. Alibaba Cloud — Model Studio pricing (Qwen3.7 Max list price $2.50/$7.50, 1M context)
  12. DeepInfra — DeepSeek V4 Pro pricing guide
  13. GMI Cloud — DeepSeek V4 benchmarks (~1.6T parameters, 1M context)
  14. CostGoat — Kimi API pricing (K2.6)
  15. OpenRouter — Kimi K2 family (262K context)
  16. TechTimes — Meta opens Muse Spark 1.1 paid API ($1.25/$4.25 per 1M tokens)
  17. Artificial Analysis — throughput and latency benchmarks (speed figures)
  18. Artificial Analysis — Opus 4.8 vs Gemini 3.1 Pro (1M context confirmation)
  19. CodingFleet — Opus 4.8 vs GPT-5.5 (verbosity: ~110M vs ~35M avg tokens on AA Intelligence Index; 2.9× on DeepSWE; 3.35× in Composio tests; context windows)
  20. Artificial Analysis — DeepSeek V4 Pro vs GPT-5.5 (output tokens per task methodology, speeds, 1.6T/49B active parameters)
  21. DeepSeek — API pricing docs (peak-hour standard rate, effective Aug 16 2026: $1.32 input / $0.044 cached / $3.96 output per 1M)
  22. Artificial Analysis — Muse Spark 1.1 (Intelligence Index standing vs. Kimi K2.6)
  23. Artificial Analysis — Claude Opus 5 (Intelligence Index #1 at 61; ~100M output tokens on evaluation vs. ~63M median; 1M context)
  24. Artificial Analysis — GPT-5.5 (~72M output tokens on evaluation vs. ~63M median; Intelligence Index 55)
  25. OpenRouter — GPT-5.6 Sol (1M context)
  26. Artificial Analysis — GPT-5.6 Sol (Intelligence Index 61; ~81 t/s; ~70M output tokens on evaluation vs. ~66M median)
  27. OpenRouter — Kimi K3 ($3/$0.30 cached/$15 per 1M, 1M context)
  28. Artificial Analysis — Kimi K3 (Intelligence Index v4.3: 44; $3/$0.30 cached/$15 per 1M; ~37 t/s; ~160M output tokens on evaluation)
  29. Artificial Analysis — Muse Spark 1.2 (Intelligence Index 57; $1.25/$4.25 per 1M, $0.15 cached)
  30. Forbes — Alibaba's Qwen3.8-Max prices frontier AI at $2/$6 per 1M tokens
  31. MarkTechPost — Qwen3.8-Max (2.4T total / 95B active MoE, 1M context)
  32. Artificial Analysis — Qwen3.8-Max (Intelligence Index v4.3: 45; ~39 t/s; ~190M output tokens on evaluation vs. GPT-5.6 Sol’s ~90M — 2.1× thinking overhead)
  33. 9to5Google — Google launches Gemini 3.6 Flash ($1.50/$7.50 per 1M, 1M context)
  34. Artificial Analysis — Gemini 3.6 Flash (Intelligence Index 52; ~224 t/s; ~59M output tokens on evaluation vs. ~70M median)
  35. Tech Times — Google cuts Gemini 3.7 Flash price in half, launches August 2026
  36. Artificial Analysis — Gemini 3.7 Flash (Intelligence Index 56, +4 over Gemini 3.6 Flash; $0.75/$3.75 per 1M, 90% cache discount; 1M context; ~297 t/s; ~64M output tokens on evaluation vs. ~72M median)
  37. xAI — Grok 4.7 model page ($2/$0.50 cached/$6 per 1M; 500K context; higher rates above 200K input)
  38. Artificial Analysis — Grok 4.7 (Intelligence Index v4.3.2: 46, #21 of 211; $2/$6 per 1M; 500K context; ~87 t/s at xhigh; ~240M output tokens on evaluation vs. GPT-5.6 Sol’s ~90M — 2.7× thinking overhead)
  39. SpaceXAI — Introducing Grok 4.7 (Sept 21, 2026; flagship for code and agentic work, replaces Grok 4.6 on the models page)
  40. OfficeChai — DeepSeek V4 Pro 0813 benchmarks (Intelligence Index 53; 3.6× price increase over prior V4 Pro)
  41. Artificial Analysis — Gemini 3.1 Pro Preview (~134 t/s; ~67M output tokens on Intelligence Index v4.3 vs. GPT-5.6 Sol's ~90M — 0.7× thinking overhead)
  42. Artificial Analysis — DeepSeek V4 Pro 0813 model page (~90 t/s at max; ~160M output tokens on Intelligence Index v4.3 vs. GPT-5.6 Sol’s ~90M — 1.8× thinking overhead)
  43. Artificial Analysis — Claude Fable 5 (~67 t/s; ~83M output tokens on evaluation vs. GPT-5.6 Sol's ~70M — 1.2× thinking overhead)
  44. OpenAI — GPT-5.6 Sol model page (promotional pricing: $4/$0.40 cached/$20 per 1M, effective through at least Nov 21, 2026; standard list price is $5/$0.50/$30)
  45. OpenAI Developer Community — 20% price reduction for GPT-5.6 Sol API
  46. Anthropic — Introducing Claude Fable 5.1 and Claude Mythos 5.1 ($10/$50 per 1M; cache reads at 0.025× input = $0.25)
  47. Artificial Analysis — Claude Fable 5.1 (Intelligence Index v4.3: 53; ~69 t/s at max; ~190M output tokens on evaluation vs. GPT-5.6 Sol’s ~90M — 2.1× thinking overhead; 1M context)
  48. Anthropic — Introducing Claude Opus 5.5 (Sept 22, 2026; $4/$20 per 1M; cache reads at 0.05× input = $0.20)
  49. Artificial Analysis — Claude Opus 5.5 (Intelligence Index v4.3: 58, #1; ~96 t/s at max effort; max effort used ~260M output tokens on evaluation vs. GPT-5.6 Sol’s ~90M — 2.9× thinking overhead; 1M context)
  50. Artificial Analysis — LLM leaderboard, Nov 2, 2026 snapshot (median output speeds: Claude Opus 5.5 ~96 t/s at max effort; GPT-6.1 Sol ~89 t/s at max; Gemini 3.8 Flash ~137 t/s at high; DeepSeek V4 Pro 0813 ~90 t/s at max)
  51. OpenAI — GPT-6.1 Sol model page ($2/$0.10 cached/$10 per 1M; 1.05M context)
  52. ExplainX — GPT-6.1 Sol launch at DevDay, Sept 29, 2026 (same $2/$10 as GPT-6 Sol, cached input halved to $0.10)
  53. Artificial Analysis — GPT-6.1 Sol (Intelligence Index: 52, #11 of 224; ~89 t/s at max; ~67M output tokens on evaluation vs. GPT-5.6 Sol’s ~90M — 0.7× thinking overhead)
  54. Artificial Analysis — GPT-5.6 Sol, the thinking-overhead reference point (Intelligence Index v4.3: 47; ~90M output tokens on evaluation = 1.0×)
  55. 9to5Google — Gemini 3.8 Flash rolls out (Sept 2, 2026; same $0.75/$3.75 per 1M as 3.7 Flash through Dec 31, 2026)
  56. Artificial Analysis — Gemini 3.8 Flash (Intelligence Index v4.3: 41; ~137 t/s at high; ~170M output tokens on evaluation vs. GPT-5.6 Sol’s ~90M — 1.9× thinking overhead; 1M context)
  57. Artificial Analysis — Muse Spark 1.3 (Intelligence Index v4.3: 48, #24 of 219; $1.25/$0.15 cached/$4.25 per 1M; 1M context; ~127 t/s; ~170M output tokens on evaluation vs. GPT-5.6 Sol’s ~90M — 1.9×)
  58. Meta — Muse API pricing and rate limits (Muse Spark 1.3 standard mode: $1.25 input / $0.15 cached / $4.25 output per 1M; no long-context premium)
  59. OpenAI — GPT-6 Astra model page ($10/$1 cached/$50 per 1M; 1.05M context; prompts above 272K input tokens bill at 2× input and cache rates and 1.5× output)
  60. OpenAI — GPT-6 Astra: A new generation of intelligence (Sept 2026 launch; Standard API pricing $10/$50 per 1M; OpenAI’s flagship for computer use, coding, science and professional work)
  61. Artificial Analysis — GPT-6 Astra (Intelligence Index v4.3.2: 53, #7 of 217; ~71 t/s at max; ~60M output tokens on evaluation vs. GPT-5.6 Sol’s ~90M — 0.7× thinking overhead; 1M context)
  62. Google — Gemini 4 Argon: Built to solve your hardest problems (Oct 13, 2026; introductory $2/$10 per 1M with cached input 95% off, through Nov 30, 2026; $4/$20 from Dec 1, 2026; 1M unified input-output token limit)
  63. Artificial Analysis — Gemini 4 Argon (Intelligence Index v4.3.2: 53, #8 of 217; ~110M output tokens on evaluation vs. GPT-5.6 Sol’s ~90M — 1.2× thinking overhead; 1M context; ~84 t/s at high)
  64. OpenRouter — Gemini 4 Argon ($2/$0.10 cached/$10 per 1M introductory; P50 throughput, Oct 19, 2026: ~31 t/s on Google AI Studio, ~45 t/s on Google Vertex)