List prices as of 8 Oct 2026
Claude Haiku 5.5 now lists at $0.10 per 1M input tokens, a tenth of Haiku 4.5's $1. The cheap tier of every big lab is now very cheap. Here is one cheap, one mid and one frontier model from each lab, what each one is for, and how to work out what your job will cost before you pick.
The cheat-sheet (USD per 1M tokens, input / output)
Standard API list prices for short prompts, read on 8 Oct 2026. Prices change often, so check the page before you commit.
CHEAP TIER
- Claude Haiku 5.5: $0.10 in / $0.50 out. Classifying, tagging, extraction, routing, subagents.
- GPT-6 Luna: $0.10 in / $0.50 out. High-volume chat, tagging, summaries.
- Gemini 3.5 Flash-Lite: $0.30 in / $2.50 out. High-volume agent tasks, translation, simple data processing.
MID TIER
- Claude Sonnet 5.5: $2.00 in / $10.00 out. Everyday writing, coding and analysis.
- GPT-6.1 Sol: $2.00 in / $10.00 out. Everyday writing, coding and analysis.
- Gemini 3.8 Flash: $0.75 in / $3.75 out. Coding, agents and tool use (intro price to 31 Dec 2026).
FRONTIER TIER
- Claude Fable 5.1: $10.00 in / $50.00 out. Hard reasoning, long-running agent work.
- GPT-6 Astra: $10.00 in / $50.00 out. Hard reasoning, the toughest tasks.
- Gemini 3.1 Pro (Preview): $2.00 in / $12.00 out. Hard multimodal reasoning, agentic coding.
The rule of thumb: start every job on the cheap tier. Move up one tier only when the output fails your check. The frontier tier costs up to 100x the cheap tier (Haiku 5.5 to Fable 5.1, GPT-6 Luna to GPT-6 Astra).
The fine print that changes the bill
- Claude Haiku 5.5 is priced by prompt length: prompts over 100,000 tokens pay $0.50 / $2.50. Claude Opus 5.5 sits between Sonnet and Fable at $4 / $20. Haiku 4.5 is still listed at $1 / $5.
- OpenAI lists a short-context and a long-context price. GPT-6 Luna long context: $0.20 / $0.75. GPT-6.1 Sol: $4 / $15. GPT-6 Astra: $20 / $75.
- Gemini 3.8 Flash is $0.75 / $3.75 through 31 Dec 2026, then $1.50 / $7.50 from 1 Jan 2027. Gemini 3.1 Pro is $4 / $18 for prompts over 200K tokens. Gemini output prices include thinking tokens.
- Reasoning or thinking tokens are billed as output on all three. A model that thinks a lot can cost more than its price per token suggests.
- Batch: all three labs list batch prices at half the standard price (for example Haiku 5.5 $0.05 / $0.25, GPT-6 Luna $0.05 / $0.25, Gemini 3.1 Pro $1 / $6).
- Caching: input you send again and again (a long system prompt, the same document) is billed at a fraction of the input price when it is cached. Each page lists the cached price.
The cost formula
Cost = (tokens in ÷ 1,000,000) × input price + (tokens out ÷ 1,000,000) × output price
A token is about 3/4 of an English word, so 1,000 words is roughly 1,300 tokens.
Worked example: summarize 10,000 support emails a month
Each email is about 1,500 tokens in, and each summary is about 200 tokens out.
- Tokens in: 10,000 × 1,500 = 15M
- Tokens out: 10,000 × 200 = 2M
Cost a month:
- Claude Haiku 5.5: 15 × $0.10 + 2 × $0.50 = $2.50
- GPT-6 Luna: 15 × $0.10 + 2 × $0.50 = $2.50
- Gemini 3.5 Flash-Lite: 15 × $0.30 + 2 × $2.50 = $9.50
- Gemini 3.8 Flash: 15 × $0.75 + 2 × $3.75 = $18.75
- Claude Sonnet 5.5: 15 × $2 + 2 × $10 = $50
- GPT-6.1 Sol: 15 × $2 + 2 × $10 = $50
- Gemini 3.1 Pro: 15 × $2 + 2 × $12 = $54
- Claude Fable 5.1: 15 × $10 + 2 × $50 = $250
- GPT-6 Astra: 15 × $10 + 2 × $50 = $250
Same job, from $2.50 to $250 a month. For a summary, the cheap tier is usually enough: test 20 real emails on it first.
Copy it into a sheet
Tokens in go in column A and tokens out in column B, from row 2. Put the input price in E1 and the output price in F1.
=A2/1000000*$E$1 + B2/1000000*$F$1
Pick in 3 steps
- Write down the job and how you will check the output (for example: "the summary names the customer's problem in one line").
- Run 20 real examples on the cheap tier of the lab you already use. If 19 of 20 pass your check, stop here.
- Move up one tier only for the examples that failed. Many teams route: cheap model first, a bigger model only when the cheap one is unsure.
The official pricing pages (check before you commit)
- Anthropic (Claude): https://docs.anthropic.com/en/docs/about-claude/pricing
- OpenAI: https://platform.openai.com/docs/pricing
- Google (Gemini API): https://ai.google.dev/gemini-api/docs/pricing
All prices above are list prices in USD read from these pages on 8 Oct 2026. They are not quotes, and they change. Check the page on the day you build.
Researching a competitor or a market? CompEdge turns a company name into a sourced brief, where every figure shows its source and period: https://compedge.nayrix.com/samples