Direct answer · checked November 8, 2026
The cheapest models tie closely on price
For a cached coding-agent turn (a 60K-token prompt with 80% cache reads and 2K output tokens), the cheapest model on list price is Gemini 2.5 Flash-Lite at $0.00248 per request. Claude Haiku 5.5 and GPT-6 Luna tie next at $0.00268.
One tier up, Claude Sonnet 5.5, GPT-6.1 Sol and Gemini 4 Argon all list $2 input, $0.10 cache read and $10 output, so they tie at $0.0488 per request. They split apart elsewhere: Gemini 4 Argon doubles on December 1, 2026, and GPT-6.1 Sol charges more for prompts over 272K tokens.
Compare API list prices on your workload
List prices per 1M tokens, checked November 8, 2026 on each vendor's pricing page. Enter one request's token mix to add a monthly cost column. Everything runs in your browser.
Load a workload
Cheapest of 31 shown: Gemini 2.5 Flash-Lite at $24.80 a month ($0.00248 per request).
| $/1M | $/1M | $/1M | $/1M | tokens | per month | Notes | ||
|---|---|---|---|---|---|---|---|---|
| Gemini 2.5 Flash-Lite#1 cheapestgemini-2.5-flash-lite · older | $0.10 | $0.01 | —no separate write rate listed | $0.40 | 1,048,576 | $24.80$0.00248/req | Cache storage $1.00 per 1M tokens per hour. | |
| Claude Haiku 5.5#2 cheapestclaude-haiku-5-5 · current | Anthropic | $0.10 | $0.01 | $0.125 | $0.50 | 1,000,000 | $26.80$0.00268/req | Prompts over 100,000 tokens pay $0.50 input / $2.50 output. Prompt length includes cache reads and writes. |
| GPT-6 Luna#2 cheapestgpt-6-luna · current | OpenAI | $0.10 | $0.01 | $0.125 | $0.50 | 1,050,000 | $26.80$0.00268/req | Over 272K input tokens: $0.20 input / $0.75 output. |
| GPT-5.6 Lunagpt-5.6-luna · older | OpenAI | $0.20 | $0.02 | $0.25 | $1.20 | 1,050,000 | $57.60$0.00576/req | Over 272K input tokens: $0.40 input / $1.80 output. |
| Gemini 3.1 Flash-Litegemini-3.1-flash-lite · current | $0.25 | $0.025 | —no separate write rate listed | $1.50 | 1,048,576 | $72.00$0.0072/req | Cache storage $1.00 per 1M tokens per hour. | |
| Gemini 3.5 Flash-Litegemini-3.5-flash-lite · current | $0.30 | $0.03 | —no separate write rate listed | $2.50 | 1,048,576 | $100.40$0.01004/req | Cache storage $1.00 per 1M tokens per hour. | |
| Gemini 3 Flash Previewgemini-3-flash-preview · preview | $0.50 | $0.05 | —no separate write rate listed | $3.00 | 1,048,576 | $144.00$0.0144/req | Google calls it a legacy Flash model. Cache storage $1.00 per 1M tokens per hour. | |
| Gemini 3.8 Flashgemini-3.8-flash · current | $0.75 | $0.075 | —no separate write rate listed | $3.75 | 1,048,576 | $201.00$0.0201/req | Price through December 31, 2026; $1.50 input / $7.50 output from January 1, 2027. Cache storage $0.50 per 1M tokens per hour. | |
| Gemini 3.6 Flashgemini-3.6-flash · older | $0.75 | $0.075 | —no separate write rate listed | $3.75 | 1,048,576 | $201.00$0.0201/req | Previous-generation Flash, same price as 3.8 Flash. Price through December 31, 2026; $1.50 input / $7.50 output from January 1, 2027. | |
| GPT-5.4 minigpt-5.4-mini · older | OpenAI | $0.75 | $0.075 | —no separate write rate listed | $4.50 | 400,000 | $216.00$0.0216/req | 400K context window. No rate listed for prompts over 272K tokens. |
| GPT-6.1 Terragpt-6.1-terra · current | OpenAI | $1.00 | $0.05 | $1.25 | $5.00 | 1,050,000 | $244.00$0.0244/req | Over 272K input tokens: $2 input / $7.50 output. |
| Claude Haiku 4.5claude-haiku-4-5 · older | Anthropic | $1.00 | $0.10 | $1.25 | $5.00 | 200,000 | $268.00$0.0268/req | Legacy model; previous tokenizer. 200K context window. |
| Claude Sonnet 5.5claude-sonnet-5-5 · current | Anthropic | $2.00 | $0.10 | $2.50 | $10.00 | 1,000,000 | $488.00$0.0488/req | Cache reads cost 0.05x input. Same price across the full 1M window. |
| GPT-6.1 Solgpt-6.1-sol · current | OpenAI | $2.00 | $0.10 | $2.50 | $10.00 | 1,050,000 | $488.00$0.0488/req | Over 272K input tokens: $4 input / $15 output. |
| Gemini 4 Argongemini-4-argon · current | $2.00 | $0.10 | —no separate write rate listed | $10.00 | 1,048,576 | $488.00$0.0488/req | Price through November 30, 2026; $4 input / $20 output from December 1, 2026. Implicit caching only. | |
| Claude Sonnet 5claude-sonnet-5 · older | Anthropic | $2.00 | $0.20 | $2.50 | $10.00 | 1,000,000 | $536.00$0.0536/req | Legacy model. The $2 / $10 launch price is now the standard price. |
| GPT-6 Solgpt-6-sol · current | OpenAI | $2.00 | $0.20 | $2.50 | $10.00 | 1,050,000 | $536.00$0.0536/req | Over 272K input tokens: $4 input / $15 output. |
| GPT-5.2gpt-5.2 · older | OpenAI | $1.75 | $0.175 | —no separate write rate listed | $14.00 | 400,000 | $574.00$0.0574/req | 400K context window. No rate listed for prompts over 272K tokens. |
| GPT-5.6 Terragpt-5.6-terra · older | OpenAI | $2.00 | $0.20 | $2.50 | $12.00 | 1,050,000 | $576.00$0.0576/req | Over 272K input tokens: $4 input / $18 output. |
| Gemini 3.1 Pro Previewgemini-3.1-pro-preview · preview | $2.00 | $0.20 | —no separate write rate listed | $12.00 | 1,048,576 | $576.00$0.0576/req | Prompts over 200K tokens: $4 input / $18 output. Cache storage $4.50 per 1M tokens per hour. | |
| GPT-5.4gpt-5.4 · older | OpenAI | $2.50 | $0.25 | —no separate write rate listed | $15.00 | 1,050,000 | $720.00$0.0720/req | No separate cache-write charge listed. Over 272K input tokens: $5 input / $22.50 output. |
| Claude Sonnet 4.6claude-sonnet-4-6 · older | Anthropic | $3.00 | $0.30 | $3.75 | $15.00 | 1,000,000 | $804.00$0.0804/req | Legacy model; previous tokenizer. |
| Claude Opus 5.5claude-opus-5-5 · current | Anthropic | $4.00 | $0.20 | $5.00 | $20.00 | 1,000,000 | $976.00$0.0976/req | Cache reads cost 0.05x input. Fast mode: $8 input / $40 output. |
| GPT-5.6 Solgpt-5.6-sol · older | OpenAI | $4.00 | $0.40 | $5.00 | $20.00 | 1,050,000 | $1,072$0.1072/req | Promotional pricing runs until November 30, 2026. Over 272K input tokens: $8 input / $30 output. |
| Claude Opus 5claude-opus-5 · older | Anthropic | $5.00 | $0.50 | $6.25 | $25.00 | 1,000,000 | $1,340$0.1340/req | Legacy model. Fast mode: $10 input / $50 output. |
| Claude Opus 4.8claude-opus-4-8 · older | Anthropic | $5.00 | $0.50 | $6.25 | $25.00 | 1,000,000 | $1,340$0.1340/req | Legacy model. |
| Claude Opus 4.7claude-opus-4-7 · older | Anthropic | $5.00 | $0.50 | $6.25 | $25.00 | 1,000,000 | $1,340$0.1340/req | Legacy model; first model on the newer tokenizer. |
| GPT-5.5gpt-5.5 · older | OpenAI | $5.00 | $0.50 | —no separate write rate listed | $30.00 | 1,050,000 | $1,440$0.1440/req | No separate cache-write charge listed. Over 272K input tokens: $10 input / $45 output. |
| Claude Fable 5.1claude-fable-5-1 · current | Anthropic | $10.00 | $0.25 | $12.50 | $50.00 | 1,000,000 | $2,320$0.2320/req | Cache reads cost 0.025x input. Same price across the full 1M window. |
| Claude Fable 5claude-fable-5 · older | Anthropic | $10.00 | $1.00 | $12.50 | $50.00 | 1,000,000 | $2,680$0.2680/req | Legacy model; Fable 5.1 lists cheaper cache reads. |
| GPT-6 Astragpt-6-astra · current | OpenAI | $10.00 | $1.00 | $12.50 | $50.00 | 1,050,000 | $2,680$0.2680/req | Over 272K input tokens: $20 input / $75 output. |
Anthropic: Cache write shown is the 5-minute write (1.25x input); a 1-hour write costs 2x input. US-only inference (inference_geo: "us") adds 1.1x on Claude 4.6 and later models. Batch API: 50% off input and output. Pricing page, checked November 8, 2026.
OpenAI: Cache write shown is the 30-minute write where OpenAI lists one; a 12-hour write costs 2.5x input. Regional processing (data residency) endpoints add 10% for models released on or after March 5, 2026. Batch and Flex are cheaper; Fast and Ultrafast cost more. Pricing page, checked November 8, 2026.
Google: Paid tier, Standard. Text input price shown; audio input costs more on some models. Explicit context caching also bills storage per 1M tokens per hour; there is no per-write charge. Output price includes thinking tokens. Pricing page, checked November 8, 2026.
Transparent math
How the workload column is calculated
The cache-read share splits each prompt into cached and uncached input. The prompt size picks the rate card: Haiku 5.5 switches above 100,000 tokens, Gemini 3.1 Pro Preview above 200K, and the OpenAI models with long-context rates above 272K. Where a vendor lists no cache-read rate, cached tokens bill at the input rate.
Fair comparison
How to compare LLM API prices fairly
Price cache reads, not just input
Coding agents resend the same context every turn, so most input is often a cache hit. Cache-read rates vary more than input rates. Claude Fable 5.1 and GPT-6 Astra both list $10 input, but a cache read is $0.25 on Fable 5.1 and $1 on Astra. Set the cache-read share from your own logs.
Count cache writes separately
Anthropic charges 1.25x input for a 5-minute cache write and 2x for a 1-hour write. OpenAI lists a 30-minute write on its newer models and a 12-hour write at 2.5x input. Gemini bills cache storage per hour rather than per write. The workload column leaves writes out, so check the write column if your prefix changes often.
Token counts differ by tokenizer
The same text is a different number of tokens on each vendor's tokenizer. Anthropic says Claude 4.7 and later models use a newer tokenizer that produces about 30% more tokens for the same text than the previous tokenizer, which Claude Sonnet 4.6 and earlier use. The exact increase depends on the content. A per-token tie does not mean an equal bill. Measure the same task on each model.
Watch tiers and dated prices
Long prompts can change the whole request's rate: Claude Haiku 5.5 charges 5x for prompts over 100,000 tokens, and that count includes cache reads. Google and OpenAI also list dated price changes. Re-check any row before you sign a budget.
List prices checked November 8, 2026
Worked scenario: one coding-agent turn, 10,000 times a month
A 60,000-token prompt with 80% cache reads (12,000 uncached, 48,000 cached) and 2,000 output tokens. This is the "Coding agent turn" preset in the tool above.
| Model | Uncached input | Cache reads | Output | Per request | Per month |
|---|---|---|---|---|---|
| Claude Haiku 5.5 | $0.001200 | $0.000480 | $0.001000 | $0.002680 | $26.80 |
| Claude Sonnet 5.5 | $0.024000 | $0.004800 | $0.020000 | $0.048800 | $488.00 |
| GPT-6.1 Sol | $0.024000 | $0.004800 | $0.020000 | $0.048800 | $488.00 |
| Gemini 4 Argon | $0.024000 | $0.004800 | $0.020000 | $0.048800 | $488.00 |
| Claude Opus 5.5 | $0.048000 | $0.009600 | $0.040000 | $0.097600 | $976.00 |
Same mix, after December 1, 2026: Gemini 4 Argon's listed rates double to $4 input, $0.20 cache read and $20 output, so this workload goes from $488 to $976 a month, the same as Claude Opus 5.5 today.
Same mix, bigger prompt: on the "Long-context repo review" preset (300K prompt, 90% cache reads, 4K output, 1,000 requests), Claude Sonnet 5.5 and Gemini 4 Argon cost $127 a month. GPT-6.1 Sol costs $234 because its over-272K rates apply.
Tokenizer check: if your token counts came from Claude Sonnet 4.6 logs, Anthropic's roughly 30% figure for the newer tokenizer would put the Sonnet 5.5 row near $634 rather than $488. Measure before you decide.
Use your own token mix
The presets are illustrations. The whoburnedmore CLI command npx whoburnedmore reads local logs from supported coding agents, signs you in and syncs daily token counts by model: input, output, cache write and cache read. Those totals give you real numbers for the workload panel. Add --local to stay offline and write a local dashboard instead. To test cache savings on one prompt, try the prompt caching savings calculator.
Frequently asked questions
What is the cheapest LLM API for coding?
On list price, checked November 8, 2026, Gemini 2.5 Flash-Lite is the cheapest model in the table for a cached coding-agent turn: $0.00248 per request for a 60K-token prompt with 80% cache reads and 2K output tokens. Claude Haiku 5.5 and GPT-6 Luna tie next at $0.00268. Price is not quality. Run the same task on each candidate and compare the bill and the result.
Is Claude more expensive than GPT or Gemini?
It depends on the model. Claude Sonnet 5.5, GPT-6.1 Sol and Gemini 4 Argon all list $2 input and $10 output per 1M tokens, and $0.10 for a cache read. Claude Opus 5.5 is $4 / $20. Claude Fable 5.1 and GPT-6 Astra are both $10 / $50, but Fable 5.1's cache read is $0.25 against Astra's $1. Claude models on the newer tokenizer can also count more tokens for the same text.
Why are Claude Code, Codex, Copilot and Cursor plans not in the table?
Subscriptions are not priced per token. A Claude Code plan, Codex, GitHub Copilot or Cursor charges a monthly fee with usage limits or credits, so a per-1M-token table can't compare them. Use the AI coding plan calculator to compare a flat plan with API-equivalent cost.
Why are some models missing?
The table lists models the vendors still offer on their standard API. Models that OpenAI's model pages mark deprecated (GPT-5.3-Codex, GPT-5.4 nano, GPT-5.1, GPT-5, GPT-5 mini and GPT-5 nano) are left out, and so are the Pro models. Google marks Gemini 2.5 Pro and 2.5 Flash deprecated; from December 1, 2026 their requests route to Gemini 3.8 Flash and 3.5 Flash-Lite and bill at those rates. Claude Mythos is limited access. xAI and DeepSeek are not included because their pricing pages could not be checked on November 8, 2026.
Do these prices include batch, fast mode or regional pricing?
No. Every rate is the standard, global, on-demand list price. Batch is 50% off at Anthropic, and OpenAI and Google list cheaper Batch and Flex tiers. Fast or priority modes cost more. US-only inference on Claude 4.6 and later, and OpenAI regional processing for models released on or after March 5, 2026, add 10%.
Which prices change soon?
Gemini 4 Argon's $2 / $10 price runs through November 30, 2026, then becomes $4 / $20. Gemini 3.8 Flash and 3.6 Flash go from $0.75 / $3.75 to $1.50 / $7.50 on January 1, 2027. OpenAI says GPT-5.6 Sol's promotional pricing runs until November 30, 2026. Check the vendor page again before you commit a budget.
Sources
- Anthropic: Claude API pricing
Base input, 5-minute and 1-hour cache writes, cache reads and output per model; Haiku 5.5's over-100,000-token prompt tier; the newer tokenizer note for Claude 4.7 and later. Checked November 8, 2026.
- Anthropic: Claude models overview
Context windows for current Claude models; legacy model pages give the rest. Checked November 8, 2026.
- OpenAI: API pricing
Standard input, cached input, 30-minute cache write and output per model, the over-272K-token long-context rates, and the GPT-5.6 Sol promotion end date. Checked November 8, 2026.
- OpenAI: model pages
Context windows and deprecation labels for each GPT model listed. Checked November 8, 2026.
- Google: Gemini API pricing
Paid-tier Standard input, context caching and output per model, the over-200K-token tiers, and the dated Gemini 4 Argon and Gemini 3.8 Flash price changes. Checked November 8, 2026.
- Google: Gemini API model pages
Input token limits and context windows for each Gemini model listed. Checked November 8, 2026.