AI API Pricing 2026: GPT-6, Claude 5.5 & Gemini 4 Cost Per Million Tokens

By Gia Gray · Updated November 2026 · 8 min read

The frontier moved again. The models people priced a year ago — GPT-4o, Claude 3.5 Sonnet, Gemini 1.5 — are now legacy, retired, or shut down, and the current generation (OpenAI's GPT-6 line, Anthropic's Claude 5.5 line, and Google's Gemini 4 Argon plus the Gemini 3.x Flash tiers) has reshuffled the cost picture. This page is the single, current per-million-token reference I keep coming back to. Every number below was checked against the provider's official pricing page on November 1, 2026; standard (non-batch) rates are shown, for Gemini's tiered models the figure is the rate for prompts up to 200K tokens, and for Claude Haiku 5.5 it's the rate for prompts up to 100K tokens.

Current AI API pricing — all major models (per 1M tokens)

Sources: OpenAI, Anthropic, Google, DeepSeek and Alibaba Qwen pricing pages, November 2026. DeepSeek rates shown are peak-hour prices (off-peak is half price); Qwen rates are Alibaba's international (Singapore) prices. Always confirm before making production decisions — these rates change often. (Promotional pricing: Gemini 4 Argon is $2/$10 through Nov 30, 2026, rising to $4/$20 on Dec 1; Gemini 3.8 Flash is $0.75/$3.75 through Dec 31, 2026, rising to $1.50/$7.50 on Jan 1, 2027.)

The three flagship tiers, and what they cost

Premium reasoning. GPT-6 Astra ($10/$50), Claude Fable 5.1 ($10/$50), Claude Opus 5.5 ($4/$20), and Gemini 4 Argon ($2/$10 promo, $4/$20 from Dec 1) sit at the top. Output is where these get expensive — Astra's and Fable's $50/M output is five times their input rate, so an output-heavy workload on a premium model is the fastest way to a surprise bill. Reach for these only when the task genuinely needs frontier reasoning, coding, or analysis.

Workhorse. GPT-6.1 Sol ($2/$10) and Claude Sonnet 5.5 ($2/$10) are the models most production apps should default to — now priced identically, with GPT-6.1 Terra ($1/$5) one step down for cost-sensitive work. Both handle the large majority of real coding, writing, and analysis work at a fraction of premium output cost, with full 1M-token context windows.

Budget / high-volume. GPT-6 Luna ($0.10/$0.50), Claude Haiku 5.5 ($0.10/$0.50), Gemini 3.8 Flash ($0.75/$3.75), and Gemini 3.5 Flash-Lite ($0.30/$2.50) are built for classification, routing, extraction, and anything you run at scale. Running a flagship on these tasks wastes 10–50× on cost for no quality gain.

The cheapest models right now

For pure cost-per-token, it's now a tie between GPT-6 Luna and Claude Haiku 5.5 at $0.10 input / $0.50 output (Haiku's rate applies to prompts up to 100K tokens). Google's cheapest current text model is Gemini 3.5 Flash-Lite ($0.30/$2.50) — Gemini 2.5 Flash-Lite no longer appears on Google's pricing page — and the legacy GPT-4o mini ($0.15/$0.60) is still callable until its January 14, 2027 shutdown. But "cheapest per token" rarely equals "cheapest per result" — a smarter model that solves the task in one pass can beat a cheaper one that needs three attempts. Benchmark on your actual task before committing.

The Chinese providers: DeepSeek and Qwen

Two Chinese labs now publish English-language API pricing that undercuts most US mid-tier models. DeepSeek bills by time of day: DeepSeek Flash (V4.1) is $0.30/$1.20 at peak and $0.15/$0.60 off-peak, and DeepSeek V4 Pro is $1.32/$3.96 at peak and half that off-peak. Peak hours are weekdays 01:00–04:00 and 06:00–10:00 UTC, so most US working hours land in the cheaper window. Alibaba's Qwen has Qwen3.8-Max at $2/$6, which is notably cheap on output for a flagship, and Qwen3.8-Flash at $0.15/$0.47, the lowest output price we track. Before you switch, check where your requests are processed and whether that fits your data and compliance rules.

Watch the new cost driver: thinking tokens

The current generation leans heavily on reasoning ("thinking") tokens — internal tokens a model generates before its visible answer. On most current models these are billed at the output rate and can dominate the cost of a single request on hard problems. That shifts the metric that matters from cost per token to cost per correct answer: a pricier reasoning model that gets it right the first time can be cheaper end-to-end than a cheap model you have to re-run. Budget for thinking tokens explicitly on reasoning-heavy workloads.

Prompt caching still pays for itself

OpenAI, Anthropic and Google all price cached input at roughly 2.5–10% of the base input rate (a 90%+ discount on the cached portion). If your app re-sends a large, stable system prompt, document, or conversation history on every request, caching is the single biggest lever you have — often a 50–70% cut to total input cost with zero quality change. Batch APIs add another ~50% off for work that doesn't need a real-time response.

What changed recently

If you're updating an old cost model, the headline shifts: OpenAI's GPT-5.6 line gave way to GPT-6 (Astra / Sol / Terra / Luna), with the workhorse and budget tiers getting cheaper; Anthropic launched Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5, with Opus dropping to $4/$20 and Haiku to $0.10/$0.50; and Google launched Gemini 4 Argon as its new flagship, with Gemini 3.1 Pro still listed as a preview and Gemini 2.5 Flash-Lite gone from the pricing page. OpenAI has also scheduled GPT-4o and GPT-4o mini for API shutdown on January 14, 2027. Further back, Google discontinued Gemini 1.5 and shut down Gemini 2.0 on June 1, 2026. Net effect: frontier capability keeps getting cheaper, but reasoning/thinking tokens make worst-case request cost harder to predict.