LLM Model Pricing in August 2026: Every Major API and Open Model

The AI Stack Cost Report · August 2026 · Part 2: The Pricing Map. The meter moved again this month. This is StackSpend's monthly read on what the modern AI engineering stack actually costs. Also in this issue: the frontier briefing, the coding stack, and the image stack.

Use this when you want one current, apples-to-apples view of what large language models cost in August 2026 — across the closed-API frontier, the open-weight models that now match it, and the hosting providers that serve those open models.

The fast answer: July repriced the middle and bottom of the market. GPT-5.6 Luna fell 80% to $0.20/$1.20 and Terra 20% to $2/$12 on July 30; Claude Opus 5 arrived at $5/$25, doing Fable-5-class work at half Fable 5's price. The floor is now around $0.04–$0.30 per 1M input on open-weight models, the volume tier sits at $0.20–$1.50, and the closed frontier holds at $5–$10 input / $25–$50 output. One date to diarise: Claude Sonnet 5's introductory $2/$10 ends August 31, 2026, moving to $3/$15 on September 1.

Pricing pages have never been harder to compare, and July made it worse: two vendors changed rates mid-month, a third scheduled an increase, and the newest models bill differently depending on an effort setting you choose per request. This guide puts the closed frontier, open-weight models, and the hosts that serve them in one place, in the same units, for August 2026.

For rates between issues, the LLM API Pricing Index re-syncs published per-1M-token prices daily, and the model changelog logs releases and deprecations month by month. For what these choices mean after deployment — the actual bill, not the rate card — see AI cost monitoring.

Quick answer

For an August 2026 snapshot:

• Closed frontier APIs run $5–$10 per 1M input and $25–$50 per 1M output: Claude Fable 5 and Mythos 5 at $10/$50, Claude Opus 5 and GPT-5.6 Sol at $5/$25–$30.
• The volume tier got dramatically cheaper. GPT-5.6 Luna is now $0.20/$1.20 after an 80% cut, and Terra $2/$12 after a 20% cut. Gemini 3.5 Flash-Lite is $0.30/$2.50.
• Open-weight models hosted on specialist providers are still 5–50x cheaper per token than the closed frontier — but the newest frontier-scale open models (Kimi K3 at $3/$15) are priced like mid-tier closed models, not like commodity inference.
• The same open model costs different amounts on different hosts. DeepSeek V4 Pro is roughly $1.30/$2.60 on DeepInfra, $1.74/$3.48 on Fireworks, and ~$2.10 input on Together AI — a 60%+ spread on identical weights.
• A scheduled increase is coming. Claude Sonnet 5 moves from $2/$10 to $3/$15 on September 1, 2026.

How to read LLM pricing in 2026

Almost all text LLMs are still priced per million tokens, split into input (the prompt you send) and output (the response generated). A token is roughly 0.75 English words, so ~1,300 tokens per 1,000 words. Four rules now carry most of the cost math:

1. Output is priced higher than input — usually 3–6x. Output-heavy workloads (long generations, agents) are dominated by the output rate; retrieval-heavy RAG is dominated by the input rate.
2. Context size can change the rate, not just the volume. Gemini 3.1 Pro moves from $2/$12 to $4/$18 above 200K input tokens. Anthropic's Claude 4.6-and-later models, by contrast, now include the full 1M context at standard pricing. See long-context pricing above 200K tokens.
3. Caching and batch are real discounts. Batch APIs cut 50% at OpenAI, Anthropic, and Google; cache hits cost 10% of base input at Anthropic and up to 90% less at Google. On some hosts these stack.
4. New in 2026: the effort dial and the tokenizer. Claude Opus 5 exposes five effort levels, so the same prompt can cost several times more or less depending on a request parameter. And Claude 4.7-and-later models use a tokenizer that produces roughly 30% more tokens for the same text (Anthropic pricing docs) — so a per-token comparison against an older model understates the real per-request cost.

The formula that turns any rate card into a monthly estimate:

Monthly cost = ((input_tokens/1,000,000 × input_rate)
             +  (output_tokens/1,000,000 × output_rate))
             ×  requests_per_month

Closed frontier APIs (August 2026)

These are the managed, proprietary-weight APIs. Rates are list prices checked on 2 August 2026.

Mistral rates are market-quoted ranges; validate the exact SKU and deployment path at checkout. Batch pricing is 50% off input and output at OpenAI, Anthropic, and Google.

How to think about the closed tier in August. The flagship band ($5 input) is now crowded — GPT-5.6 Sol, GPT-5.5, and Claude Opus 5 all sit there, with Opus 5 the cheapest on output at $25. Above them, the $10/$50 Fable/Mythos tier is a genuinely separate purchase, and after Opus 5's release it is harder to justify for anything but the longest-horizon autonomous work. Below them, the volume tier collapsed: Luna at $0.20/$1.20 is now cheaper than several open-weight models served on managed hosts. For the head-to-head, see OpenAI vs Anthropic pricing.

Open-weight models (August 2026)

Open-weight models remain the default for high-volume work, but July split the category in two. Commodity open models still sit at pennies per million tokens; the new frontier-scale open models are priced like mid-tier closed APIs, because someone still has to serve 2.8 trillion parameters.

Rates are representative served-API prices from public host pricing pages; open-weight is not always OSI open-source, so check the licence before you ship. See closed vs open AI models in 2026 for the portfolio framing.

The thing that changed in July: an open licence no longer implies a low price. Kimi K3 at ~$3/$15 is more expensive than GPT-5.6 Terra ($2/$12) and 15x more expensive than GPT-5.6 Luna. The value of frontier open weights is control, auditability, and the option to self-host — not, at that scale, cheapness.

Open-model hosting providers (August 2026)

This is the part most pricing guides skip. The same open weights cost different amounts depending on who serves them — because you are renting that provider's hardware, throughput, and margin.

The one number that matters most: the price of the same model varies wildly by host. DeepSeek V4 Pro is roughly $1.30/$2.60 on DeepInfra, $1.74/$3.48 on Fireworks, and ~$2.10 input on Together AI — identical weights, a 60%+ spread. That is not a quality difference; it is hardware, throughput, and margin. And the cheapest per-token host is not automatically the best value: Groq and Cerebras charge more but deliver far higher throughput, which can lower total cost for latency-bound or agentic workloads that would otherwise stall.

How to choose a hosting provider

• Optimizing pure per-token cost at scale? DeepInfra (owns its stack, no aggregator margin) is the usual price leader; Together AI has the broadest catalog.
• Latency- or throughput-bound (chat, agents, tool loops)? Groq and Cerebras win on speed, which often beats a lower per-token rate on total cost.
• Want one API across many hosts with failover? OpenRouter or Hugging Face Inference Providers — a small fee for portability.
• Need fine-tuning, compliance, or dedicated capacity? Fireworks (tuning, HIPAA, speed tiers), Together AI (LoRA), or Baseten (dedicated deployments).
• GPU-heavy or non-text models? Replicate's compute-time model fits better than per-token.
• Enterprise governance already on a hyperscaler? Route open models through Bedrock or Vertex Model Garden — see Hugging Face vs direct provider APIs.

What this costs in practice

Illustrative example. A content-moderation classifier: 400-token input, 50-token output, 2M requests/month. Rates are the August 2026 list prices above; your real bill will differ with retries, caching, and prompt drift.

Two things to notice. First, the same task spans 13x across reasonable choices. Second — and this is new this month — GPT-5.6 Luna at $280 now beats a managed open-model host on several of last month's comparisons. Before the July 30 cut, the same Luna row would have been $1,400. A price change you did not action is a saving you did not take.

The pricing traps to watch

• Silent repricing. Two of the tiers in this guide changed price on July 30 with no action required from you. If nobody compares spend-per-model month over month, a cut looks identical to no change.
• Scheduled increases. Claude Sonnet 5 goes from $2/$10 to $3/$15 on September 1, 2026. Batch goes from $1/$5 to $1.50/$7.50 the same day.
• Effort settings. Claude Opus 5's five-level effort dial changes token consumption per request by multiples. The rate card is the same; the bill is not.
• Tokenizer drift. Claude 4.7-and-later models emit roughly 30% more tokens for the same text than earlier Claude models — a per-token comparison across generations is not like-for-like.
• Long-context cliffs. Gemini 3.1 Pro doubles input pricing above 200K tokens. RAG pipelines that grow their retrieved context drift over the cliff without anyone changing a model.
• Same-model host drift. Routing a model through a different host (or letting an aggregator re-route) can change your effective rate 60%+ with no code change.
• Idle GPU on dedicated endpoints. Hugging Face Endpoints, Baseten dedicated, and self-hosting bill for capacity, not tokens — an always-on GPU costs the same at 5% utilization as at 95%.

Every one of these is invisible on a pricing page and obvious on the invoice.

FAQ

What is the cheapest way to run an LLM in August 2026?

For most high-volume text workloads the floor is around $0.02–$0.20 per 1M input: open-weight models (Llama 3.1 8B, DeepSeek V4 Flash, Qwen3-235B, GPT-OSS) on a direct-stack host like DeepInfra. What changed in July is that the cheapest closed option is now competitive with them — GPT-5.6 Luna at $0.20/$1.20 after its 80% cut, against Gemini 3.5 Flash-Lite at $0.30/$2.50. See cheapest AI API for chat, RAG, and coding.

How much does Claude Opus 5 cost?

Claude Opus 5 is $5 per 1M input tokens and $25 per 1M output — the same as Opus 4.8. Cache hits are $0.50/1M, 5-minute cache writes $6.25/1M, and Batch API pricing is $2.50/$12.50. Fast mode (research preview) is billed at $10/$50. The full 1M-token context is included at standard pricing.

When does Claude Sonnet 5's introductory pricing end?

August 31, 2026. From September 1, 2026, Claude Sonnet 5 bills at $3 input / $15 output per 1M tokens instead of the introductory $2/$10 — a 50% increase. Batch pricing moves from $1/$5 to $1.50/$7.50, and cache hits from $0.20 to $0.30 per 1M.

Why does the same open model cost different amounts on different hosts?

Because you are paying for the provider's hardware, throughput, and margin — not the weights, which are free to download. In August 2026, DeepSeek V4 Pro is roughly $1.30/$2.60 on DeepInfra, $1.74/$3.48 on Fireworks, and ~$2.10 input on Together AI. The pricier hosts often deliver higher tokens-per-second or added capability (fine-tuning, compliance), which can reduce total cost on latency-bound workloads even at a higher per-token rate.

Are open models still cheaper than closed APIs?

At the commodity end, yes — 5–50x cheaper per token, with no infrastructure to run. At the frontier end the gap has closed: Kimi K3, a 2.8T open-weight model, is quoted around $3/$15, which is more than GPT-5.6 Terra at $2/$12. Frontier open weights now buy control, auditability, and self-hosting optionality rather than a lower price. If you self-host, add GPU, autoscaling, observability, and on-call — that only pays off at steady high volume.

How do I compare LLM providers on price fairly?

Put every option in the same units ($/1M input and $/1M output), estimate your average input and output tokens per request, multiply by monthly volume, and apply any batch/cache discount you can actually use. Then validate against real spend — list prices ignore retries, tool calls, effort settings, tokenizer differences, long-context cliffs, and idle capacity. StackSpend does this continuously across providers: AI cost monitoring.

Track LLM spend across every provider

The hard part of LLM pricing in 2026 isn't any single rate card — it's that a modern stack spans a closed API, one or two open-model hosts, and sometimes an aggregator on top, each invoicing separately in different units on different cycles. The combined picture doesn't exist unless you build it. July proved the second half of the problem: rates moved twice in one month, in opposite directions, and nothing in your codebase changed.

Connect your providers to StackSpend for one view of total LLM spend with model-level breakdown, daily model recommendations that flag when a cheaper model would do the same job, and anomaly detection that catches a runaway agent the day it moves. Start with LLM cost monitoring or provider setup guides for OpenAI, Anthropic, Fireworks, and Hugging Face.

About The AI Stack Cost Report

The AI Stack Cost Report is StackSpend's monthly briefing on what the modern AI engineering stack costs — frontier and open-weight LLMs, AI coding tools, image generation, and the hosting and infrastructure they run on. Each issue is compiled from primary pricing pages and release reporting, with source-check dates listed at the top of every article. The August 2026 issue has four parts: the frontier briefing, the pricing map, the coding stack, and the image stack.

Between issues, the LLM API Pricing Index re-syncs published per-1M-token rates daily and the model changelog logs every release and deprecation by month.

A monthly snapshot tells you what the market charges; it can't tell you what your stack spent yesterday. StackSpend does — per-provider AI spend, anomaly alerts the day a spike happens, and a daily report in Slack. Setup is read-only, takes about five minutes, and starts with a free 14-day trial. Why we publish it: the meter came back.

Related reading

• The Latest in LLMs, August 2026
• LLM Model Pricing in July 2026 — the previous issue
• Current AI API Pricing (2026): OpenAI, Grok, Anthropic, Gemini
• Closed vs Open AI Models in 2026: A Practical, Balanced Guide
• Cheapest AI API in 2026 for Chat, RAG, and Coding
• Long-Context AI Pricing in 2026: What Happens Above 200K Tokens
• LLM API Pricing Index — live per-1M-token rates, updated daily

References

• Anthropic — Pricing (Claude Platform docs)
• Anthropic — Introducing Claude Opus 5
• OpenAI Pricing
• CNBC — OpenAI cuts prices for two of its GPT-5.6 AI models
• VentureBeat — AI price wars: OpenAI cuts GPT-5.6 Luna prices by 80%
• Google Vertex AI Generative Pricing
• Amazon Bedrock Pricing
• xAI Models and Pricing
• OpenRouter — Grok 4.5
• OpenRouter — Kimi K3
• Mistral Pricing
• Together AI Pricing
• Groq On-Demand Pricing
• Fireworks AI Serverless Pricing
• DeepInfra Pricing
• Hugging Face Inference Providers Pricing
• Cerebras Pricing
• ThursdAI — July 2026 AI releases