LLM API pricing comparison (all providers)

This is an LLM API price comparison across the eight providers that matter for production traffic — OpenAI, Anthropic, Google, xAI, DeepSeek, Alibaba (Qwen), Mistral, and Meta (Muse Spark) — organized by tier so you compare flagship against flagship and workhorse against workhorse. The sticker prices are one screen of numbers; the caveats underneath them are what actually decide the winner, so both are here.

All prices are USD per 1M tokens. "Cached input" is the discounted rate for reading a previously cached prompt prefix. Where a value is not published in the source dataset it is shown as "—".

Flagship tier

The frontier models, plus each provider's headline offering. Note the spread between Claude Fable 5.1 and GPT-6 Astra at the top and DeepSeek V4 Pro at the bottom — roughly 15× on input, 25× on output. These are not substitutes on quality, but the price gap is real.

Read the flagship tier as four bands: the new top pair, where GPT-6 Astra and Claude Fable 5.1 both list $10.00 / $50.00 — identical to the cent, and separate only on cache reads, where Fable 5.1 and Mythos 5.1 price hits at 0.025× input ($0.25) against Astra’s $1.00; frontier Western models ($4 input and up — Claude Opus 5.5, GPT-5.6 Sol and Meta's Muse Wonder preview, now tied at $4 / $20, and legacy Claude Opus 5), value flagships ($1.25–$2.00 input — GPT-6.1 Sol, Gemini 4 Argon, Gemini 3.1 Pro, Grok 4.7, 4.6 and 4.5, Qwen3.8-Max, Mistral Medium 3.5, Muse Spark 1.3); and DeepSeek V4 Pro alone at the bottom on price. Fable 5.1 and Mythos 5.1 arrived on 2026-09-01 as one model at two safeguard levels — Mythos 5.1 is restricted to vetted cybersecurity and life-sciences organizations, so most teams are pricing Fable 5.1. Astra arrived on 2026-09-03 and is available in the API: OpenAI's API changelog lists it as released that day, with no access restriction. Our 2026-09-04 update said API access was limited at launch — OpenAI's API documentation does not support that, so we have withdrawn it. Two more flagships landed on 2026-09-21 and 2026-09-22: xAI's Grok 4.7, at the same $2.00 / $6.00 as Grok 4.6, and Claude Opus 5.5, which Anthropic's model guide names as the starting point for capability-first work, at $4 / $20 with cache reads at $0.20 (0.05× input). OpenAI's GPT-6.1 Sol, released 2026-09-29, lists $2.00 / $10.00 with cache reads at $0.10 — level with Gemini 4 Argon's launch rate and below every other flagship here except DeepSeek V4 Pro. Google's Gemini 4 Argon, generally available since 2026-10-13, lists the same $2.00 / $10.00 and $0.10 cache reads, with a 1M window and no 200K long-context surcharge — but only until 2026-12-01, when it steps up to $4.00 / $20.00 (cache $0.20). Meta now sells Muse Spark 1.3 on its own Model API at $1.25 / $4.25, with cache reads at $0.15 and no long-context premium; a cheaper Contributor tier, where Meta may train on your prompts, is not shown. Meta's Muse Wonder, in preview, lists $4.00 / $20.00 with cache reads at $0.20 and a 1M window. The caveats section below explains why the cheapest rows come with strings attached.

Gemini vs Grok: the two value flagships

Gemini 3.1 Pro and Grok 4.5 both list $2.00 input, so head-to-head the difference is output: Grok 4.5 is $6.00 against Gemini 3.1 Pro's $12.00 — half the output cost — which makes Grok materially cheaper for generation-heavy work. Two caveats reorder that. Gemini 3.1 Pro is preview-priced and can change at general availability, and it adds a long-context surcharge above 200K tokens (input and cached input 2×, output 1.5×). Grok 4.5 doubles input, cached input and output above 200K tokens — on every token in the request — and tops out at a 500K window versus Gemini's 1M. Under 200K tokens and output-heavy, Grok wins on price; if you need the 1M window, Gemini is the one that has it. xAI's newer Grok 4.6 and Grok 4.7 (launched 2026-09-21) keep the same $2.00 / $6.00 list rates, 500K window, and 200K surcharge threshold (cached input rises to $0.50), so the comparison holds for them too.

Workhorse (mid) tier

This tier carries most production traffic — good enough for the majority of tasks, priced for volume. It is also the most competitive, with a roughly 6× spread on input alone.

Claude Sonnet 5's $2 / $10 began as an introductory price, but Anthropic made it the permanent standard rate in August 2026 — the scheduled increase to $3 / $15 was cancelled. Claude Sonnet 5.5, released 2026-09-28, lists at the same $2 / $10 with 1M context; it launched with $0.20 cache reads, and Anthropic cut those to $0.10 on 2026-10-07. Gemini 3.6 Flash is promo-priced at $0.75 / $3.75 through 2026-12-31 and lists $1.50 / $7.50 from 2027. Gemini 3.8 Flash is now Google's current Flash model, at the same rates; 3.6 Flash is the previous generation. Grok 4.3 is the standout on output at $2.50, a fraction of GPT-5.4's $15.00 — for chat and generation workloads where output dominates, it is the cheapest capable mid-tier model on a US-hosted API. OpenAI's GPT-6.1 Terra, released 2026-10-26, lists $1.00 / $5.00 with cache reads at $0.05 (0.05× input) and a 1.05M window — half GPT-5.6 Terra's input rate. Qwen3.7-Plus, Alibaba's recommended Plus model, bills thinking and non-thinking output alike at $1.60, and Alibaba currently shows a 20% limited-time discount on it; the older qwen-plus alias is $1.20 non-thinking but $4.00 in thinking mode. Both triple input and output above 256K input tokens. Correction (2026-11-02): this table previously listed Qwen-Plus cached input at $0.04, its explicit-cache read rate; the column uses each model's implicit-cache rate, which for qwen-plus is $0.08. Mistral Large 4, in public preview, is Mistral's named replacement for both Large 3 and Medium 3.5, which it deprecated on 2026-10-31 and retires on 2027-04-30; it lists $0.70 / $2.10 with cache reads at $0.07 and a 1M window. Llama 4 Maverick has left the table: Together AI removed its serverless endpoint on 2026-03-31, and it is now offered only on dedicated endpoints and for fine-tuning.

Reasoning models: you pay for the thinking

Dedicated reasoning models spend extra output tokens "thinking" before they answer, so their output rate — and how many output tokens they burn — matters more than input. OpenAI sells these as a separate line; Anthropic and Google fold extended thinking into their main models, where the thinking tokens bill at the standard output rate.

o3 is the practical reasoning workhorse that remains — OpenAI shut down o4-mini on 2026-10-23, and o3 and o3-pro retire on 2026-12-11, the last of the deprecated o-series; its deprecation table names GPT-5.6 Terra as the replacement for o4-mini and GPT-5.6 Sol for o3 and o3-pro (o3-pro in Sol's pro reasoning mode). o3-pro and GPT-5.5 Pro are maximum-effort variants whose output rates ($80–$180) make them expensive to let run long. Because reasoning models emit far more output tokens than a normal chat turn, their effective cost per task can exceed a flagship's even when the per-token rate looks moderate.

The budget tier, in one line

Below the workhorse tier sits a field of sub-dollar models — GPT-6 Luna, Claude Haiku 5.5, GPT-5.6 Luna, GPT-5.4 nano, Gemini Flash-Lite, DeepSeek V4.1 Flash, Qwen3.7-Flash, Qwen3.5-Flash, Qwen-Flash and Qwen3.8-Flash, Ministral 3 3B, Mistral Small 4, and legacy GPT-5 nano — where the cheapest options run about 35× less than the priciest "budget" model. That tier has its own ranked matrix; see the cheapest-LLM guide linked below.

The caveats that reorder the ranking

Every cheap row above carries a footnote. Read these before you rank on price:

• ◦Preview pricing: Gemini 3.1 Pro is priced as a preview and can change at general availability.
• ◦Off-peak vs peak: DeepSeek's rates are off-peak; the peak rate is 2× on both V4 Pro and V4.1 Flash, and applies only 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday, excluding Chinese public holidays — weekends and those holidays are off-peak all day. DeepSeek introduced that peak/off-peak split on 2026-08-16, then cut its budget tier on 2026-09-10 when DeepSeek-V4.1-Flash replaced V4-Flash — older figures still circulate; the table above is current.
• ◦Long-context surcharge: Gemini 3.1 Pro doubles input (cached input rises from $0.20 to $0.40) and adds 50% to output above 200K tokens; Grok 4.7, 4.6 and 4.5 double input and output above 200K, and once a prompt reaches 200K the higher rate applies to every token in the request, cached input included.
• ◦Long-context surcharge, continued: GPT-6 Astra bills 2× input and cached input and 1.5× output once a prompt passes 272K tokens, applied to the entire request — $20.00 / $75.00 (cached $2.00) rather than $10.00 / $50.00 (cached $1.00).
• ◦Qwen model IDs: Alibaba's current flagship, Qwen3.8-Max, is a flat $2.00 / $6.00 up to 1M tokens, but older IDs are still sold at different rates — the qwen-max alias ($1.60 / $6.40) has a 32K window and is updated on a rolling basis. Our 2026-09-22 update said it was pinned to a January 2025 snapshot — that was wrong: Alibaba says only that it is functionally equivalent to that snapshot, which was itself decommissioned on 2026-05-13 (re-checked 2026-11-02). Qwen3-Max, previously listed here from $1.20 / $6.00, was retired on 2026-10-10 in Alibaba's mass legacy-model retirement. Check which model ID your code actually calls (Singapore-region list prices; other regions differ).
• ◦Promotions: OpenAI cut GPT-5.6 Sol from $5.00 / $30.00 to $4.00 / $20.00 (cache $0.40) on 2026-08-21 and says the promotional rate is available at least through 2026-11-21 (still in effect, re-verified 2026-11-02); it has not published what happens after that date, so do not assume either the cut or the old price persists. Gemini 3.6 Flash bills $0.75 / $3.75 (cache $0.075) only through 2026-12-31, then $1.50 / $7.50 from 2027-01-01.
• ◦Tokenizer skew: per-token prices are not comparable across tokenizer generations — Anthropic's newest models emit roughly 30% more tokens for the same text. Compare cost per task, not price per token.

Which should you pick

Match the model to the job, not to the top of a leaderboard:

• ◦Cheapest capable flagship: DeepSeek V4 Pro off-peak ($0.66 / $1.98) still sits far below the Western flagships — if its peak-hours 2× and China-hosted API fit your constraints.
• ◦Best value flagship on a US-hosted API: Grok 4.7 ($2.00 / $6.00) for output-heavy work; Gemini 3.1 Pro ($2.00 / $12.00) when you need the 1M window and can accept preview pricing.
• ◦Frontier quality: Claude Opus 5.5 and GPT-5.6 Sol (promotional) are tied at $4.00 / $20.00 — Sol is cheaper on uncached text after tokenization, Opus 5.5 on cache-heavy work with $0.20 cache reads; Claude Fable 5.1 ($10.00 / $50.00, cache reads $0.25) when you want Anthropic's most capable model and cost is secondary.
• ◦Workhorse default: Claude Sonnet 5.5 ($2 / $10, cache reads $0.10) or promo-priced Gemini 3.6 Flash; Grok 4.3 ($1.25 / $2.50) is the cheapest of the mid tier for output-heavy traffic.

What is the cheapest LLM API?

On sticker price, DeepSeek V4 Pro off-peak ($0.66 / $1.98) is the cheapest capable flagship, and in the budget tier Qwen3.7-Flash ($0.030 / $0.130, for prompts under 32K tokens) is the cheapest overall. But "cheapest" depends on your input:output ratio, whether you hit peak-hours surcharges, and how your text tokenizes — price your own workload rather than trusting a single headline number.

Are the prices the same across providers for the same quality?

No. At the flagship tier the input rate ranges from $0.66 (DeepSeek V4 Pro off-peak) to $10.00 (Claude Fable 5) — about a 15× spread — and the models are not interchangeable on quality, latency, tool support, or data residency. Price is one axis; the caveats above are the others.

Why do some models cost more above a token threshold?

Serving very long contexts is more expensive, so several providers add a long-context surcharge. Gemini 3.1 Pro doubles input and cached input and adds 50% to output above 200K tokens; Grok 4.7, 4.6 and 4.5 double input and output, cached input included, on the whole request once a prompt passes 200K. If your prompts are long, the headline rate understates your bill — model the surcharge tier explicitly.

How current are these prices?

They are list prices as of 2026-11-02 and they move often — most recently Anthropic cut Claude Sonnet 5.5's cache reads from $0.20 to $0.10 on 2026-10-07; before that, OpenAI cut its GPT-5.6 Sol flagship to a promotional $4 / $20 (in effect at least through 2026-11-21), DeepSeek split its pricing into peak and off-peak bands and then cut its budget tier, Anthropic made Claude Sonnet 5's intro price permanent, xAI shipped Grok 4.6 and then Grok 4.7 at the same price, and Anthropic launched Claude Opus 5.5 at $4 / $20, 20% below Opus 5. Re-verify against each provider before you commit a budget.