AI-NAV · Article
AI Model Pricing 2026: ChatGPT, Claude, Gemini & DeepSeek
The AI market now spans hundreds of models across three distinct families — text (chat) models, reasoning models, and image models — each with a fundamentally different pricing logic. A text model bills per token; a reasoning model bills per token but also charges for invisible "thinking" tokens; an image model bills per image or per megapixel. This guide lays out the real numbers across the global leaders.
1. The pricing logic: three families, three meters
| Family | Billing unit | Key cost trap |
|---|---|---|
| Text / Chat | Per 1M input + output tokens | Output tokens cost 3–8× input; cache hits can cut input cost 90% |
| Reasoning | Per 1M tokens including hidden thinking tokens | A 500-token answer may consume 3,000+ hidden reasoning tokens |
| Image | Per image / per megapixel | Resolution tiers (1K/2K/4K) and "editing steps" multiply per-image cost |
Iternal AI LLM pricing calculator and vendor pricing pages (Oct 2026).
2. Global flagship text & reasoning models
Closed-model leaderboard, October 2026. List rates, USD per 1M tokens (uncached input / output). The market clusters into four clear price bands — premium at $50 output, flagship at $20, workhorses at $6–12, and small models at $0.50. Gemini 3.8 Flash's intro rate sits between bands as a transitional tier, joining the workhorse band in January 2027.
| Model | Provider | Input / 1M | Output / 1M | Context | Positioning |
|---|---|---|---|---|---|
| Claude Fable 5.1 | Anthropic | $10.00 | $50.00 | 1M | Premium flagship |
| GPT-6 Astra | OpenAI | $10.00 | $50.00 | 1.1M | Premium flagship |
| Claude Opus 5.5 | Anthropic | $4.00 | $20.00 | 1M | Frontier default |
| GPT-6.1 Sol | OpenAI | $2.00 | $10.00 | 1.1M | Mid-tier default |
| Claude Sonnet 5.5 | Anthropic | $2.00 | $10.00 | 1M | Coding & agents |
| Gemini 3.1 Pro | $2.00 | $12.00 | 500K | Multimodal reasoning | |
| Grok 4.7 | xAI | $2.00 | $6.00 | 500K | Value flagship |
| Gemini 3.8 Flash | $0.75 | $3.75 | 1M | Intro rate (→$1.50/$7.50 Jan 2027) | |
| GPT-6 Luna | OpenAI | $0.10 | $0.50 | 1.1M | Cheapest small model |
| Claude Haiku 5.5 | Anthropic | $0.10 | $0.50 | 1M | Cheapest small model |
*Sources: Perch AI "Open vs closed AI model pricing" (Oct 2026), AI Vanguard (Oct 11, 2026), Builders Camp (Oct 2026).*
The reasoning-model premium
Reasoning models add "thinking tokens" that are billed but never shown:
| Reasoning model | Provider | Input / 1M | Output / 1M | Note |
|---|---|---|---|---|
| o3 | OpenAI | $2.00 | $8.00 | Current; launched at ~$10/$40 |
| Gemini 2.5 Pro | $1.25 | $10.00 | Multimodal strength | |
| DeepSeek R1 | DeepSeek | $0.55 | $2.19 | ~1/4 of o3's current price (~1/18 vs launch) |
*Sources: AI Tool Analysis "DeepSeek R1 vs o3 vs Gemini 2.5 Pro" (May 2026), StackSpend pricing index (Oct 2026).*
3. Open-weight & Chinese models: the price disruptors
USD equivalents use ~7.2 CNY/USD (October 2026).
| Model | Lab | Input / 1M | Output / 1M | Note |
|---|---|---|---|---|
| DeepSeek V4.1 Flash | DeepSeek | $0.15–0.30 | $0.60–1.20 | Off-peak to peak; CN reseller ~¥1/¥2 |
| DeepSeek V4 Pro | DeepSeek | $1.32 | $3.96 | Peak; off-peak half (CN ~¥12/¥24) |
| Qwen 3.8 Flash | Alibaba | $0.11 | $0.38 | ¥0.8 / ¥2.7 |
| Qwen 3.8 Max | Alibaba | $1.67 | $5.00 | ¥12 / ¥36 |
| GLM 5.3 Flash | Zhipu / Z.ai | $0.15 | $0.50 | Official API list |
| GLM 5.3 | Zhipu / Z.ai | $1.40 | $4.40 | Official API list |
| Kimi K3 | Moonshot | $3.00 | $15.00 | ¥20/¥100; output ~50× DeepSeek Flash tier (¥2) |
| MiniMax M3 | MiniMax | $0.30 | $1.20 | ¥2.1 / ¥8.4 |
| Kimi K2.7 Code | Moonshot | $0.95 | $4.00 | ¥6.5 / ¥27 |
*Sources: Perch AI (open-weight table, Oct 2026), 尧图网络 (Aug 2026), 阿里云百炼官方定价页, 算力圈 (Oct 2026). Note: DeepSeek announced up to 3× price rises in mid-2026; Kimi K3 remains the most expensive domestic flagship per output token. Mixed peak/off-peak — verify live billing pages before committing.*
4. Image models: per-image, per-megapixel, and subscriptions
Two economic models: subscription (Midjourney, ChatGPT, Gemini app) and API per-image (Imagen, Gemini image, FLUX, Stable Diffusion).
| Model / Tool | Channel | Price | Positioning |
|---|---|---|---|
| FLUX.2 Klein 4B | API | ~$0.014 / image | Cheapest API; Apache 2.0 |
| Imagen 4 Fast | Vertex AI | ~$0.02 / image | High-volume cheap |
| GPT Image 2 | API | ~$0.03 / image | Text rendering |
| Nano Banana 2.1 | Gemini API | $0.0336 (1K) / $0.113 (4K) | Google's recommended workhorse |
| Nano Banana 2 | Gemini API | $0.045–$0.151 | Deprecated Oct 2026 |
| Gemini 3 Pro Image | Gemini API | $0.134 (1K/2K) / $0.24 (4K) | Premium, reference-image editing |
| Midjourney v7.1 | Subscription | $10–$120 / month | Art direction ceiling |
| ChatGPT (GPT Image) | Subscription | $20 / month Plus | All-in-one workflow |
| Adobe Firefly | Subscription | $9.99 / month | Commercial indemnification |
| Stable Diffusion 4 | Self-host | Free + GPU cost | Full control, open weights |
*Sources: Google Gemini API pricing page (Oct 2026), AI Tool Analysis (May 2026), AI Flash Report (Oct 2026), LaoZhang AI (Oct 2026).*
5. Subscriptions vs API: the consumer half of the market
API token rates are only half the picture. Most non-developer users buy monthly plans, and the industry has quietly standardized on ~$20/month as the "serious user" price point — ChatGPT Plus, Claude Pro, Google AI Pro, Perplexity Pro and Copilot all sit within a dollar of it. The real divergence is in the power tiers, which now stretch to $300/month.
| Plan | Provider | Price / month | What you get |
|---|---|---|---|
| ChatGPT Go / Plus / Pro | OpenAI | $8 / $20 / $100–200 | Plus: GPT-5.6 Terra + Luna; Pro: 5–20× limits, top models |
| Claude Pro / Max | Anthropic | $20 / $100–200 | Pro: Sonnet + Opus; Max: 5–20× usage, Fable included |
| Google AI Plus / Pro / Ultra | $4.99 / $19.99 / $99.99–199.99 | Pro: Gemini Pro + 5 TB storage; Ultra: Deep Think | |
| SuperGrok / Heavy | xAI | $30 / $300 | Frontier Grok access, X integration |
| Perplexity Pro / Max | Perplexity | $20 / $200 | Cited answers; Max for heavy research |
| Microsoft 365 Premium | Microsoft | ~$20 | Copilot inside Word / Excel / Outlook |
| GLM Coding Lite / Pro / Max | Zhipu | $18 / $80 / $168 | Coding-agent prompt quotas |
| Midjourney | Midjourney | $10–120 | Image tiers; Standard $30 is the creator sweet spot |
Team seats cluster at $25–30: ChatGPT Business $25, Claude Team $25, Grok Business $30 per seat.
*Sources: aipricing.guru subscription index (Sep 2026), aipricecompare.org (Aug 2026), Notebookcheck AI subscription comparison (Jul 2026).*
Subscription or API: the break-even question
The two meters suit two different workloads. If you are a human chatting all day, the subscription wins by a wide margin — a heavy Plus user would burn far more than $20 at list API rates. If you are software making bursty calls, the API wins — you pay nothing at idle, and routing 70% of traffic to budget models drops the blended bill below any flat plan. The crossover is roughly "dozens of long conversations a day": below it, API is cheaper; above it, subscribe.
Which plan for which job
- Research with verifiable citations: Perplexity Pro.
- Coding and long documents: Claude Pro — the consensus developer pick.
- Deep in the Google ecosystem: Google AI Pro (bundles storage + Gmail/Docs integration).
- All-rounder, not sure yet: ChatGPT Plus, the broadest tool set.
- Office-heavy workflows: Microsoft 365 Premium, AI inside Word and Excel.
- Real-time X and news: SuperGrok.
- Image-heavy creators: Midjourney Standard at $30.
- In China: DeepSeek, Qwen and Kimi web apps are largely free, and domestic memberships undercut every US plan on this list.
6. The bottom line: how to actually compare
Five rules that survive every price change:
- Never compare list prices alone. Tokenizers differ (Anthropic's new tokenizer produces ~30% more tokens for the same text), so measure your real prompt on each model.
- Reasoning tokens are a hidden multiplier. Budget 3–5× your expected output for reasoning models.
- Cache or pay 10×. Prompt caching cuts input cost 70–90% on every major provider; batch APIs cut 50% for non-real-time work.
- Route, don't overpay. 70% of queries can go to a budget model; only ~10% need a frontier model. Blended cost drops 60–80%.
- Prices move fast. Gemini 3.8 Flash doubles on Jan 1 2027; DeepSeek already raised up to 3×. Build your budget on a live pricing page, not a saved article.
The single most important number in this guide: the output-price spread between the most and least expensive capable models is roughly 100× — and that order of magnitude is not new. A year ago, Claude Opus 4.1 ($75 output) vs GPT-4o mini ($0.60) was already 125×. What actually changed is the capability you get at each price point: a year ago $0.60 bought a mini-tier model; today's $0.50 small models handle most everyday tasks. Prices were never flattened — they have always been stratified, which means the winners are the teams that match each task to the right tier, not the ones that default to a single "best" model.
Primary sources (all accessed October 2026): Perch AI "Open vs closed AI model pricing"; AI Vanguard "Best AI Models in 2026" pricing snapshot (Oct 11); Builders Camp LLM cost guide; Iternal AI token-cost calculator; StackSpend LLM API Pricing Index; Google Gemini API pricing page; AI Tool Analysis image-pricing breakdown; AI Flash Report image generators; aipricing.guru subscription index; aipricecompare.org; Notebookcheck AI subscription comparison; 阿里云百炼官方定价页; 算力圈; 尧图网络国产模型成本实测.