LLM API Pricing Tracker: Input and Output Prices by Model

This page tracks the published list prices of the main hosted LLM APIs, in US dollars per million tokens: input, cached input, output and context window for OpenAI, Anthropic, Google Gemini, Mistral, xAI and DeepSeek, plus the scheduled price changes those vendors have published with an effective date. It is for engineers and FinOps people who need a number for a budget or cost model and a link that proves it. Last checked: 12 November 2026.

Latest changes

• 1 Jan 2027 (scheduled): gemini-3.8-flash and gemini-3.6-flash double from $0.75 input and $3.75 output to $1.50 and $7.50. Cached input goes from $0.075 to $0.15 and cache storage from $0.50 to $1.00 per million tokens per hour (source).
• 1 Dec 2026 (scheduled): gemini-4-argon and gemini-4-carbon double from $2 input and $10 output to $4 and $20, and cached input from $0.10 to $0.20 (source). See Gemini 4 Argon and Carbon API Prices Double. On the same day, requests to gemini-2.5-pro and gemini-2.5-flash are rerouted and billed at the rates of the models that replace them (source).
• 30 Nov 2026 (scheduled): OpenAI's pricing page says the promotional price for gpt-5.6-sol ($4 input, $20 output) is available until this date. It does not say what the price is afterwards (source).
• 12 Nov 2026: Google released gemini-4-hydrogen as generally available at $1.50 input and $7.50 output, with a free tier (source).
• 10 Nov 2026: Google released gemini-4-carbon as generally available at an introductory $2 input and $10 output through 30 November (source).
• 9 Nov 2026: Anthropic launched Claude Fable 5.2 and Claude Mythos 5.2 at $8 input and $40 output. Fable 5.1 and Mythos 5.1 stay listed at $10 and $50 (source). See Claude Fable 5.2 Launches at $8/$40.
• 26 Oct 2026: OpenAI released gpt-6.1-terra (source), listed at $1 input, $0.05 cached input and $5 output (source). See OpenAI GPT-6.1 Terra.
• 7 Oct 2026: Anthropic launched Claude Haiku 5.5 and cut prompt cache reads on Claude Sonnet 5.5 from $0.20 to $0.10 per million tokens (source).
• 6 Oct 2026: Mistral Large 4 entered public preview with launch pricing of 50% off for two weeks (source). Its model page now marks it generally available at $0.70 input and $2.10 output (source). See Mistral Large 4 Open Weights.
• 29 Sep 2026: OpenAI released gpt-6.1-sol at $2 input, $0.10 cached input and $10 output for prompts up to 272K input tokens (source). See OpenAI DevDay 2026.

Flagship and mid-tier model prices (USD per 1M tokens)

Standard tier, short-context rates where a vendor has two, as in force on 12 November 2026.

Things the table cannot show:

• Claude Mythos 5.2 has the same prices as Fable 5.2 but is limited to organisations verified through Anthropic's programmes.
• Three of the four Gemini rows are temporary prices. See the scheduled changes table before you budget past 30 November.
• Gemini 4 models support implicit caching only. On gemini-3.8-flash, explicit context caching also bills storage at $0.50 per million tokens per hour.
• DeepSeek's peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday, excluding Chinese public holidays. Off-peak is half the peak rate.
• "Cached input" is a read price. Anthropic and OpenAI also charge for cache writes: on Fable 5.2 a 5-minute write is $10 and a 1-hour write $16 per million tokens, and on gpt-6-astra a 30-minute write is $12.50 and a 12-hour write $25.

Small and low-cost model prices (USD per 1M tokens)

The Gemini Flash-Lite input prices are for text, image and video. Audio input costs more on gemini-3.1-flash-lite ($0.50) and gemini-2.5-flash-lite ($0.30). Explicit caching on all three Flash-Lite models adds $1.00 per million tokens per hour of storage. The cheapest text model on xAI's models page is grok-build-0.1 at $1.00 input and $2.00 output, so there is no xAI row here.

Long-context surcharges

These models bill the whole request at a higher rate once the prompt crosses a threshold. The other rows above list a single rate across the full context window.

Scheduled price changes

Google's changelog lists breaking API changes for both reroutes, so test before 1 December.

How this page is maintained

Every number comes from a vendor's own pricing page, model page or changelog, fetched on the "Last checked" date, and each row links to the page it was read from. I do not use aggregators, resellers or news reports as price sources. Deliberately excluded: cloud marketplace prices (Bedrock, Vertex AI, Azure), batch, flex, fast and priority tiers, data residency uplifts, image, audio and embedding models, restricted-access models, negotiated enterprise discounts, and most older models that vendors still list. Cohere is not in the tables because its public pricing page currently lists per-token prices only for legacy Command and Aya models. Corrections go to hello@ajayk.sh.

FAQ

Which LLM API has the lowest price per million tokens right now?

By list price in the tables above, the lowest input prices are $0.10 per million on gpt-6-luna, Claude Haiku 5.5 (for prompts up to 100,000 tokens) and gemini-2.5-flash-lite, and the lowest output price is $0.15 on Ministral 3 8B. Per-token price is not cost per task: tokenizers differ, and Anthropic's pricing page says its newer tokenizer produces roughly 30% more tokens for the same text than its older one.

Is the Gemini API getting more expensive?

For four models, yes, on dates Google has already published. gemini-4-argon and gemini-4-carbon go from $2 and $10 to $4 and $20 per million input and output tokens on 1 December 2026, and gemini-3.8-flash and gemini-3.6-flash go from $0.75 and $3.75 to $1.50 and $7.50 on 1 January 2027. gemini-4-hydrogen, released on 12 November at $1.50 and $7.50, has no scheduled change on the pricing page.

How much cheaper is cached input than normal input?

Anthropic's cache reads cost 0.1x base input on most models, 0.05x on Opus 5.5 and Sonnet 5.5, and 0.025x on Fable 5.2. OpenAI lists cached input at a tenth of the input price on gpt-6-astra and gpt-6-luna and a twentieth on gpt-6.1-sol and gpt-6.1-terra. The highest ratio in these tables is xAI's grok-4.7, where cached input is a quarter of the input price.

Related: LLM Tokens and Pricing Explained (2026): $0.10 to $50 per Million, How to Cut LLM API Costs: Prompt Caching, Batching, Routing (2026) and Prompt Caching Explained (2026): 0.05x Reads, 5-Minute TTLs, Real Math.