AI Model API Pricing Compared (November 2026): Every Major Model, per Token and per Task

By stackcone · November 2026

LLM API prices now span 80x: from $0.10 per million input tokens (Claude Haiku 5.5, GPT-6 Luna) to $8 (GPT-6.2 Astra, Claude Fable 5.2). Two of the biggest models cut prices this month, two Gemini models double on December 1, and Gemini 3.8 Flash doubles on January 1.

This page lists every major model’s price in one place, then shows why the per-token price is only half the story.

Most expensive (in / out)
$8 / $40
Cheapest useful
$0.10 / $0.50
Price rises Dec 1
Gemini 4 Carbon, Argon
Price cuts in Nov
Fable 5.2, GPT-6.2 Astra

The full price table

Standard paid rates, USD per 1 million tokens, as of November 18, 2026. AA = Artificial Analysis Intelligence Index (higher is smarter).

ModelMakerInputOutputCached inputContextAA Index
Frontier
GPT-6.2 AstraOpenAI$8.00$40.00$0.40~1M#2
Claude Fable 5.2Anthropic$8.00$40.00$0.201M65 (#1)
GPT-6 Astra (previous)OpenAI$10.00$50.00$1.001.05M57
Claude Opus 5.5Anthropic$4.00$20.00—1M59
Gemini 4 CarbonGoogle$2.00 › $4.00$10.00 › $20.00Implicit1.05M60
Gemini 4 ArgonGoogle$2.00 › $4.00$10.00 › $20.00Implicit1.05M50
Mid-tier
GPT-6.2 SolOpenAI$2.00$10.00$0.10~1M57
Claude Sonnet 5.5Anthropic$2.00$10.00—1M55
Gemini 4 HydrogenGoogle$1.50$7.50$0.0751.05M51–54
GLM-5.3Z.ai$1.40$4.40——32
GPT-6.1 TerraOpenAI$1.00$5.00—1.05M51
Budget
Gemini 3.8 FlashGoogle$0.75 › $1.50 (Jan 1)$3.75 › $7.50———
DeepSeek V4.1-FlashDeepSeek$0.15 off-peak / $0.30 peak$0.60 / $1.20—1M33
MiniMax M3.1 FlashMiniMax$0.15$0.60——29
GLM-5.3-FlashZ.ai$0.15$0.50———
Claude Haiku 5.5Anthropic$0.10 (≤100K) / $0.50$0.50 / $2.50—1M36
GPT-6 LunaOpenAI$0.10$0.50—1.05M28

Arrows show scheduled changes: Gemini 4 Carbon and Argon on December 1, 2026; Gemini 3.8 Flash on January 1, 2027. “—” means not published in our sources. Always confirm on the provider’s pricing page before you commit a budget.

Price per task, not per token

Smarter models often finish in fewer steps; verbose models burn output tokens. Artificial Analysis measures what it costs to run each model through its whole test suite. Average cost per index task:

Cost per Artificial Analysis index task, as reported in November 2026 roundups.

Look at Sonnet 5.5 and GPT-6.1 Sol: same list price ($2/$10), same AA score (55), but Sol cost $1.27 per task against Sonnet’s $10.02, because Sol used far fewer tokens. That is the single most important number on this page. GPT-6.2 Sol, which replaced 6.1 Sol at the same price, scores two points higher.

OpenAI models hold most of Artificial Analysis’s cost-per-task frontier. Fable 5.2 is the most expensive per task but also the highest scoring; one hands-on test found it spent 56% more per task than Fable 5.1 despite the lower token price.

Discounts that change the math

LeverTypical savingWhere
Prompt caching90–98% off repeated inputFable 5.2 $0.20 vs $8; GPT-6.2 Astra $0.40 vs $8; Hydrogen $0.075 vs $1.50
Batch API50%Claude Batch (Fable 5.2 $4 / $20); Gemini Batch and Flex (half price)
Off-peak pricing50%DeepSeek V4.1-Flash
Lower effort / reasoningVaries by task; test itFable 5.2 at High matched Opus 5.5’s best FrontierCode score at about half Opus’s cost per task (Anthropic)
Short-context tier80%Haiku 5.5 under 100K tokens ($0.10 vs $0.50)
Surcharges to watch+10% to +100%Claude US-only inference 1.1x; Fable fast mode $16 / $80

What real workloads cost

Monthly cost at list prices, no caching:

Workload (per month)Haiku 5.5 / LunaHydrogenSol / SonnetOpus 5.5Fable 5.2 / Astra
Support bot: 100K chats × 3K in / 400 out$50$750$1,000$2,000$4,000
RAG search: 1M queries × 6K in / 300 out$750$11,250$15,000$30,000$60,000
Coding agent: 2K tasks × 400K in / 40K out$120$1,800$2,400$4,800$9,600

Haiku figures assume the ≤100K-token tier. The coding row assumes 400K input tokens across a whole multi-step task; with caching on repo context, input cost falls 90% or more, and output tokens dominate. For a RAG-specific breakdown, see our economical LLMs for RAG guide.

flowchart TD classDef q fill:#f1f5f9,stroke:#64748b,color:#0f172a classDef v fill:#ccfbf1,stroke:#0d9488,color:#115e59 classDef m fill:#dbeafe,stroke:#2563eb,color:#1e3a8a classDef f fill:#ffedd5,stroke:#ea580c,color:#7c2d12 S["Request"]:::q --> Q1{"Classify, extract,\nroute or tag?"}:::q Q1 -->|yes| B["Budget tier\n$0.10 to $0.75 input"]:::v Q1 -->|no| Q2{"Answer, draft,\neveryday code?"}:::q Q2 -->|yes| M["Mid tier\n$1.50 to $2 input"]:::m Q2 -->|no| F["Frontier tier\n$4 to $8 input"]:::f

Most production traffic belongs in the bottom two tiers

Five ways to cut your bill this month

  1. Move off Carbon/Argon before December 1 if Hydrogen at xhigh passes your evals. That is a 2.7x input saving after the increase.
  2. Turn on caching for system prompts, tool definitions and document context.
  3. Batch anything that can wait an hour: reports, enrichment, evals.
  4. Route by difficulty. A cheap model handles most requests; escalate the rest. Our routing guide shows a no-classifier approach.
  5. Measure cost per successful task in your own logs, not price per token.

FAQ

What is the cheapest AI model API in November 2026?

Claude Haiku 5.5 (under 100K tokens) and GPT-6 Luna, both at $0.10 per million input tokens and $0.50 per million output tokens. DeepSeek V4.1-Flash costs $0.15/$0.60 off-peak.

How much does the GPT-6.2 API cost?

GPT-6.2 Astra costs $8 per million input tokens and $40 per million output tokens, with cached input at $0.40. GPT-6.2 Sol costs $2/$10 with cached input at $0.10.

How much does the Claude API cost?

Claude Fable 5.2 is $8/$40 per million tokens, Opus 5.5 $4/$20, Sonnet 5.5 $2/$10, and Haiku 5.5 $0.10/$0.50 up to 100K tokens ($0.50/$2.50 above). Batch halves Fable 5.2 to $4/$20.

How much does the Gemini API cost?

Gemini 4 Hydrogen is $1.50/$7.50 per million tokens. Gemini 4 Carbon and Argon are $2/$10 until November 30, 2026, then $4/$20. Gemini 3.8 Flash is $0.75/$3.75 until December 31, then $1.50/$7.50.

Is a cheaper model always cheaper to run?

No. Claude Sonnet 5.5 and GPT-6.1 Sol had the same $2/$10 list price and the same AA Index score, but Sol cost $1.27 per index task against Sonnet’s $10.02 because it used far fewer tokens. Measure cost per completed task.

Sources

Prices as published by providers and reported November 9–18, 2026. Prices change often; confirm on each provider’s pricing page.

AI bill growing faster than usage?

stackcone audits LLM spend and builds routing, caching and batch pipelines that cut cost per task without hurting quality.