llm-api-pricing-reference

LLM API Pricing Reference — September 2026

Current published API rates for the major LLM providers, with the pricing tiers and thresholds that most comparison tables leave out. Every figure below was read off the provider’s own pricing page on 13 September 2026. No third-party trackers.

USD per 1,000,000 tokens unless stated otherwise.


The short version

Four things break a simple price comparison, and almost nobody shows them:

  1. Gemini Pro and Grok 4.6 change rate above 200,000 prompt tokens. The published price doubles.
  2. OpenAI’s published rates cover context under 270,000 tokens. There is no published rate above it.
  3. Claude 4.6 and later have no context tier at all — one rate across the full 1M window.
  4. DeepSeek prices by time of day, not by context length.

OpenAI

Model Input Cached input Output
GPT-5.6 Sol $5.00 $0.50 $30.00
GPT-5.6 Terra $2.00 $0.20 $12.00
GPT-5.6 Luna $0.20 $0.02 $1.20

Threshold: the pricing page states these rates apply to context lengths under 270,000 tokens. No rate is published above that.


Anthropic (Claude)

Model Input Output Cache write (5m) Cache read
Claude Fable 5.1 $10.00 $50.00 $12.50 $0.25
Claude Mythos 5.1 $10.00 $50.00 $12.50 $0.25
Claude Opus 5 $5.00 $25.00 $6.25 $0.50
Claude Sonnet 5 $2.00 $10.00 $2.50 $0.20
Claude Sonnet 4.6 $3.00 $15.00 $3.75 $0.30
Claude Haiku 4.5 $1.00 $5.00 $1.25 $0.10

No long-context tier. Claude 4.6 and later include the full 1M-token context window at standard pricing: a 900,000-token request bills at the same per-token rate as a 9,000-token one.

Cache reads are not a flat 10% everywhere. Fable 5.1 and Mythos 5.1 read at 0.025× base input ($0.25); every other model uses the 0.1× multiplier.

Fast mode (Opus 5 and 4.8 only): $10 input / $50 output, across the full window.


Google (Gemini)

Model Input Output Cache read Notes
Gemini 3.1 Pro (Preview) ≤200K $2.00 $12.00 $0.20 doubles above 200K
Gemini 2.5 Pro ≤200K $1.25 $10.00 $0.125 doubles above 200K
Gemini 3.8 / 3.7 / 3.6 Flash $0.75 $3.75 $0.075 introductory rate
Gemini 3.5 Flash $1.50 $9.00 $0.15  
Gemini 3.5 Flash-Lite $0.30 $2.50 not available no context caching

Threshold: above 200,000 prompt tokens, both Pro models bill at double the rates above.

Expiry: the 3.x Flash rates are introductory and run through 31 December 2026. They increase on 1 January 2027.

Status: Gemini 3.1 Pro is still Preview. Gemini 2.5 Pro is the only generally available Pro-tier model.

Service tiers: Batch and Flex are 50% of Standard; Priority is roughly 80% above it. Flex is synchronous, so it does not carry Batch’s asynchronous wait.


xAI (Grok)

Model Input Cached input Output
Grok 4.6 ≤200K prompt tokens $2.00 $0.50 $6.00
Grok 4.6 above 200K $4.00 $1.00 $12.00

Server-side tools are billed separately and are missing from every comparison table: Web/X search $5 per 1,000 calls, X user profiles $10/1K, file attachments $10/1K, code execution $5/1K, collections search $2.50/1K, plus per-GiB storage and egress.


DeepSeek

Model Off-peak input Off-peak output Peak input Peak output
DeepSeek V4-Pro $0.66 $1.98 $1.32 $3.96
DeepSeek V4-Flash $0.22 $0.66 $0.44 $1.32

No context tier — a time-of-day tier instead. Peak runs 01:00–04:00 and 06:00–10:00 UTC, seven hours of every twenty-four. Traffic that straddles both windows pays a blend, so a single headline rate is not a real number for DeepSeek.

Cache hits discount input by roughly 30×.


Where the comparison actually gets decided

Price per token decides less than it looks. A model at half the rate that needs three times the output tokens to finish the same job is not cheaper.

The LLM Waves leaderboards track intelligence, coding and agentic indices, output speed, context window and Arena Elo across 323 models, and the economics board ranks by cost per task rather than by headline rate. What is measured and what is estimated is set out in the methodology.


Corrections

Rates move, and this file will go stale. If a figure here no longer matches the provider’s own pricing page, open an issue or a PR with the link — vendor documentation only, not third-party trackers.

License

MIT.