LLM API prices change every few weeks, differ by tier, prompt length and time of day, and are published on a dozen separate provider pages, so most cost estimates quote a figure that is already out of date. This tracker collects the per-token list prices that model providers publish on their own pricing pages and model cards, in one dated set of tables. It covers standard, cached and batch rates, long-context thresholds, time-of-day pricing and announced price changes. It is written for engineering leads, FinOps and procurement teams, and anyone building a cost model for an LLM application. It does not explain how token billing works; that is in Model Pricing Explained, which we do not repeat here. Providers are listed alphabetically, and the order is never a ranking.
Last checked: 2026-11-15. Every price below was read from the provider's own page on 15 November 2026. Prices are list prices per one million tokens (MTok) before any negotiated discount, tax or cloud marketplace mark-up.
This is an independent reference page from NHI Mgmt Group based on public provider documentation; it is not sponsored or endorsed by any provider.
Key takeaways
- Input list prices span three orders of magnitude. As of 2026-11-15 the lowest standard input rates on this page are $0.10 per MTok (Claude Haiku 5.5 short prompts, GPT-6 Luna, Muse Ember short prompts) and the highest is $20 (GPT-6 Cyber, a restricted model).
- Batch is usually half price. Anthropic, Google, Mistral AI and OpenAI's lower tier show 50% of standard rates; xAI states a 20% batch discount; Alibaba Cloud lists a 50% batch discount on some Qwen models only.
- Prompt length and time of day change the rate. Seven providers charge more above a prompt-length threshold (100K, 200K, 256K, 272K or 512K tokens), and DeepSeek charges peak and off-peak rates by UTC hour.
- Announced increases are coming. Google lists higher prices for Gemini 4 Argon from 1 December 2026 and for Gemini 3.8 Flash and 3.6 Flash from 1 January 2027.
- Per-token prices do not compare directly. Tokenisers differ between providers and sometimes between model generations; Anthropic states its newer tokeniser produces about 30% more tokens for the same text.
Status key
| Label | Meaning on this page |
|---|---|
| Current | Listed by the provider as a current or recommended model on its pricing page or model list |
| Older | Still priced and callable, but superseded by a newer model from the same provider |
| Preview | Labelled preview, beta or early access by the provider; prices and availability may change |
| Restricted | Priced publicly but available only to approved customers or programmes |
| Price change announced | The provider lists a new rate with an effective date; both rates are shown |
| Not stated by <provider> | The provider's pages checked on 2026-11-15 do not publish this figure |
Current flagship and mid-tier model prices at a glance
As of 2026-11-15. USD per MTok, standard tier, shortest prompt-length band. "Cached" is the cache hit or cached input rate. Alphabetical by provider.
| Provider | Model (API ID) | Status | Input | Cached input | Output | Source |
|---|---|---|---|---|---|---|
| Alibaba Cloud (Qwen) | Qwen3.8-Max (qwen3.8-max) | Current | $2 | Not stated by Alibaba Cloud as a rate (context cache discount applies) | $6 | Model Studio pricing |
| Alibaba Cloud (Qwen) | Qwen3.7-Plus (qwen3.7-plus) | Current | $0.40 | Not stated by Alibaba Cloud as a rate | $1.60 | Model Studio pricing |
| Anthropic | Claude Fable 5.2 (claude-fable-5-2) | Current | $8 | $0.20 | $40 | Claude pricing |
| Anthropic | Claude Opus 5.5 (claude-opus-5-5) | Current | $4 | $0.20 | $20 | Claude pricing |
| Anthropic | Claude Sonnet 5.5 (claude-sonnet-5-5) | Current | $2 | $0.10 | $10 | Claude pricing |
| DeepSeek | DeepSeek-V4-Pro-0813 (deepseek-v4-pro) | Current | $1.32 peak, $0.66 off-peak | $0.044 peak, $0.022 off-peak | $3.96 peak, $1.98 off-peak | DeepSeek pricing |
| Gemini 4 Argon (gemini-4-argon) | Current, price change announced | $2.00, then $4.00 from 2026-12-01 | $0.10, then $0.20 | $10.00, then $20.00 | Gemini API pricing | |
| Meta | Muse Wonder (muse-wonder-1.0) | Preview | $4.00 | $0.20 | $20.00 | Meta Model API pricing |
| Meta | Muse Spark (muse-spark-1.3) | Current | $1.25 | $0.15 | $4.25 | Meta Model API pricing |
| MiniMax | MiniMax-M3 | Current | $0.30 | $0.06 | $1.20 | MiniMax pay-as-you-go pricing |
| Mistral AI | Mistral Large 4 (mistral-large-4) | Current | $0.70 | $0.07 | $2.10 | Mistral Large 4 model card |
| OpenAI | GPT-6 Astra (gpt-6-astra) | Current | $10.00 | $1.00 | $50.00 | OpenAI API pricing |
| OpenAI | GPT-6.1 Sol (gpt-6.1-sol) | Current | $2.00 | $0.10 | $10.00 | OpenAI API pricing |
| OpenAI | GPT-6.1 Terra (gpt-6.1-terra) | Current | $1.00 | $0.05 | $5.00 | OpenAI API pricing |
| xAI | Grok 4.7 (grok-4.7) | Current | $2.00 | $0.50 | $6.00 | xAI pricing |
| Z.ai (Zhipu) | GLM-5.3 (glm-5.3) | Current | $1.40 | $0.26 | $4.40 | Z.ai pricing |
Cohere's public pricing page lists prices only for legacy Command and Command R models offered to existing customers; per-token prices for Command A+, Command A and Command A Reasoning are not stated by Cohere on that page.
Small and low-cost model prices
As of 2026-11-15. USD per MTok, standard tier, shortest prompt-length band. Alphabetical by provider.
| Provider | Model (API ID) | Status | Input | Cached input | Output |
|---|---|---|---|---|---|
| Alibaba Cloud (Qwen) | Qwen3.8-Flash (qwen3.8-flash) | Current | $0.15 | Not stated by Alibaba Cloud as a rate | $0.47 |
| Anthropic | Claude Haiku 5.5 (claude-haiku-5-5) | Current | $0.10 | $0.01 | $0.50 |
| DeepSeek | DeepSeek-V4.1-Flash-1017 (deepseek-flash) | Current | $0.30 peak, $0.15 off-peak | $0.006 peak, $0.003 off-peak | $1.20 peak, $0.60 off-peak |
| Gemini 3.8 Flash (gemini-3.8-flash) | Current, price change announced | $0.75, then $1.50 from 2027-01-01 | $0.075, then $0.15 | $3.75, then $7.50 | |
| Gemini 3.5 Flash-Lite (gemini-3.5-flash-lite) | Current | $0.30 | $0.03 | $2.50 | |
| Gemini 3.1 Flash-Lite (gemini-3.1-flash-lite) | Current | $0.25 (audio $0.50) | $0.025 | $1.50 | |
| Meta | Muse Ember (muse-ember-1.0) | Current | $0.10 | $0.004 | $0.50 |
| Mistral AI | Mistral Small 4 (mistral-small-2603) | Current | $0.15 | Not stated by Mistral AI on the model card | $0.60 |
| OpenAI | GPT-6 Luna (gpt-6-luna) | Current | $0.10 | $0.01 | $0.50 |
| xAI | Grok 4.3 (grok-4.3) | Older | $1.25 | $0.20 | $2.50 |
| Z.ai (Zhipu) | GLM-5.3-Flash | Current | $0.15 | $0.03 | $0.50 |
| Z.ai (Zhipu) | GLM-5.3-FlashX | Current | $0.37 | $0.075 | $1.25 |
Restricted and specialised model prices
As of 2026-11-15. Public list prices for models whose access is limited or that target one domain. USD per MTok.
| Provider | Model | Access | Input | Cached input | Output |
|---|---|---|---|---|---|
| Anthropic | Claude Mythos 5.2 | Restricted | $8 | $0.20 | $40 |
| Anthropic | Claude Mythos 5.1 | Restricted | $10 | $0.25 | $50 |
| OpenAI | GPT-6 Cyber (gpt-6-cyber) | Restricted | $20.00 | $2.00 | $100.00 |
| OpenAI | GPT-6 Rosalind (gpt-6-rosalind) | Restricted | $15.00 | $1.50 | $75.00 |
| OpenAI | GPT-6 Astra for Law (gpt-6-astra-law) | Specialised; promotional pricing stated as available at least through 2026-12-31 | $10.00 | $1.00 | $50.00 |
| xAI | Grok Build 0.1 (grok-build-0.1) | Preview (early access) | $1.00 | $0.20 | $2.00 |
Batch, flex and priority tiers
As of 2026-11-15. How each provider prices asynchronous and premium processing, as stated on its pricing page.
| Provider | Lower-cost tier | Premium tier |
|---|---|---|
| Alibaba Cloud (Qwen) | 50% batch discount on qwen3.6-flash; Qwen3.8 and Qwen3.7 models list context cache only; batch and cache discounts cannot be combined | Not stated by Alibaba Cloud on the page checked |
| Anthropic | Batch API: 50% off input and output (for example Opus 5.5 $2 / $10) | US-only inference (inference_geo "us") at 1.1x on Claude 4.6 and later |
| DeepSeek | Off-peak rates, half of peak, outside 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays and all day on weekends and Chinese public holidays | Not stated by DeepSeek |
| Batch and Flex tiers at 50% of standard (for example Gemini 3.5 Flash-Lite $0.15 / $1.25) | Priority tier at 1.8x standard (for example Gemini 3.5 Flash-Lite $0.54 / $4.50) | |
| Meta | Contributor mode for Muse Spark 1.3 and 1.2: $0.10 input, $0.002 cached, $0.20 output | Not stated by Meta |
| MiniMax | Standard tier prices above include a 50% permanent discount on the original price | Priority tier (service_tier "priority") at 1.5x Standard, for example MiniMax-M3 $0.45 / $1.80 |
| Mistral AI | Batch processing reduces the price by 50% | Not stated by Mistral AI |
| OpenAI | Lower tier at 50% of Standard (for example GPT-6.1 Sol $1.00 / $5.00) | Fast tier at 2x Standard; Ultrafast tier for GPT-6 Astra ($60 / $300) and GPT-6.1 Sol ($12 / $60) |
| xAI | 20% batch discount on Grok 4.x models; no batch discount on Grok Build 0.1 | Not stated by xAI |
| Z.ai (Zhipu) | Cached input storage marked "Limited-time Free" | Not stated by Z.ai |
OpenAI's pricing page shows the lower and premium tiers as unlabelled tabs; the tier names above follow the tab order and should be confirmed on the page before use in a contract.
Long-context and prompt-length pricing
As of 2026-11-15. Where a threshold applies, the higher rate applies to the whole request, not only the tokens above the threshold, unless stated otherwise.
| Provider | Model | Threshold | Rate above threshold (input / output) |
|---|---|---|---|
| Alibaba Cloud (Qwen) | Qwen3.7-Plus | Over 256K input tokens (to 1M) | $1.20 / $4.80 |
| Anthropic | Claude Haiku 5.5 | Over 100,000 prompt tokens, including cache reads and writes | $0.50 / $2.50 |
| Gemini 3.1 Pro Preview; Gemini 2.5 Pro | Over 200K prompt tokens | $4.00 / $18.00; $2.50 / $15.00 | |
| Meta | Muse Ember | Over 200K total input tokens, including cached and stored conversation | $0.30 / $1.50 |
| MiniMax | MiniMax-M3 | Over 512K input tokens | $0.60 / $2.40 |
| OpenAI | GPT-6 Luna, GPT-6.1 Terra, GPT-6 Cyber (per model pages) | Over 272K input tokens | 2x input and cached input, 1.5x output for the full request, for example GPT-6 Cyber $40.00 / $150.00 |
| xAI | All Grok 4.x and Grok Build 0.1 | Prompt of 200K tokens or more | Double the standard rate, for example Grok 4.7 $4.00 / $12.00 |
Anthropic states that Claude 4.6 and later models other than Haiku 5.5 include the full 1M-token context at standard pricing. See Context Windows Explained.
Older models still priced
| Provider | Model | Input | Output |
|---|---|---|---|
| Alibaba Cloud (Qwen) | Qwen3.7-Max; Qwen3.6-Plus (to 256K); Qwen3.6-Flash (to 256K) | $2.50; $0.50; $0.25 | $7.50; $3.00; $1.50 |
| Anthropic | Opus 5, 4.8, 4.7, 4.6, 4.5; Sonnet 5; Sonnet 4.6, 4.5; Haiku 4.5 | $5; $2; $3; $1 | $25; $10; $15; $5 |
| Gemini 3.6 Flash; Gemini 2.5 Flash; Gemini 2.5 Flash-Lite | $0.75 (then $1.50 from 2027-01-01); $0.30; $0.10 | $3.75 (then $7.50); $2.50; $0.40 | |
| MiniMax | MiniMax-M2.7; M2.5; M2.1; M2 | $0.30 | $1.20 |
| xAI | Grok 4.6; Grok 4.5; Grok 4.20 (0309) | $2.00; $2.00; $1.25 | $6.00; $6.00; $2.50 |
| Z.ai (Zhipu) | GLM-5.2, GLM-5.1; GLM-5 | $1.40; $1.00 | $4.40; $3.20 |
Retirement dates for these models are tracked in the LLM Deprecation and Retirement Tracker.
What to watch next
- Google, 1 December 2026: Gemini 4 Argon standard rates double to $4.00 input and $20.00 output; batch to $2.00 / $10.00.
- Google, 1 January 2027: Gemini 3.8 Flash and 3.6 Flash rise to $1.50 / $7.50, with context cache storage from $0.50 to $1.00 per MTok per hour.
- OpenAI, 30 November 2026: OpenAI's pricing page states: "GPT-5.6 Sol's promotional pricing is available until November 30, 2026." The price tables on the same page list the model ID gpt-6.1-sol and contain no gpt-5.6-sol row, so the page does not say which listed ID the promotional note refers to (checked 2026-11-15).
- Alibaba Cloud: Qwen3.7-Plus is shown at list price with a limited-time 20% discount; the end date was not shown on the page checked.
- Mistral AI: the mistral.ai pricing page lists "Mistral Large" at $0.5 / $1.5, while the Mistral Large 4 model card (released 2026-11-05) lists $0.7 / $2.1. Confirm which applies to your model ID before committing a budget.
- Z.ai: cached input storage is marked limited-time free and may become chargeable.
How to use this tracker
- Price the workload, not the model: multiply your measured input, cached and output tokens by the rates above. Output tokens cost two to ten times input on this page, so output-heavy agents dominate the bill.
- Check the prompt-length band: if your prompts cross 100K, 200K, 256K, 272K or 512K tokens, use the long-context table, because the higher rate applies to the whole request.
- Re-tokenise when switching: count tokens with each provider's own counter before comparing; a cheaper per-token rate can still cost more if the tokeniser produces more tokens.
- Add the access path: cloud marketplace prices (Amazon Bedrock, Google Cloud, Microsoft Foundry) can differ from first-party API prices; Anthropic states a 10% premium for regional endpoints on Bedrock and Google Cloud.
- Set controls before go-live: spend caps, per-key budgets and alerts. See How to set spend limits on LLM API keys and the Denial-of-Wallet and AI Agent Cost Runaway Guide.
Stolen API keys are billed to the key owner at these list prices, which is why LLM keys are a target. See the LLMjacking Guide.
Practitioner checklist
- Record the as-of date and source URL beside every price in your cost model.
- Model input, cached input and output separately; never use a blended single rate.
- Check whether your typical prompt crosses a long-context or prompt-length threshold.
- Route non-urgent work to batch or off-peak tiers where the provider offers them.
- Put announced price changes (Google, 1 December 2026 and 1 January 2027) into next quarter's budget now.
- Measure tokens with each provider's counter before comparing per-token prices across providers or model generations.
- Set per-key spend limits and alerts, and inventory every LLM API key that can incur charges.
- Re-check this page and the provider pages before signing commitments; prices here are list prices only.
Standards and references
- Anthropic: Claude API pricing (accessed 2026-11-15)
- OpenAI: API pricing (accessed 2026-11-15)
- Google: Gemini API pricing (accessed 2026-11-15)
- xAI: API pricing (accessed 2026-11-15)
- Alibaba Cloud: Model Studio model pricing (accessed 2026-11-15)
- DeepSeek: Models and pricing (accessed 2026-11-15)
- Meta: Pricing and rate limits (accessed 2026-11-15)
- Z.ai: Pricing (accessed 2026-11-15)
Related NHI Mgmt Group resources: Model Pricing Explained · LLM Release Tracker · Small and Low-Cost LLMs Compared · Frontier AI Models Compared · How to set spend limits on LLM API keys