What AI actually costs right now: eight models, compared on live pricing

Restrict TrustList's AI-model board to the models carrying real, current pricing - eight of them, as of 31 August 2026 - and a genuine pattern shows up: the frontier got cheap fast.

The eight, ranked by output price

Model	Provider	Input / 1M	Output / 1M	Context
Gemini 3.7 Flash	Google	$0.75	$3.75	1M
o3	OpenAI	$2.00	$8.00	200K
Gemini 2.5 Pro	Google	$1.25	$10.00	1M
GPT-4o	OpenAI	$2.50	$10.00	128K
Claude Sonnet 5	Anthropic	$2.00	$10.00	1M
Qwen3.8-Max	Alibaba	$2.00	$6.00	1M
Grok 4.6	xAI	$2.00	$6.00	500K
Claude Opus 5	Anthropic	$5.00	$25.00	1M

Seven of the eight sit within a tight $6-10 output-price band. Only Claude Opus 5 - Anthropic's actual flagship, not its mid-tier model - sits well above it, and even that is roughly half what a top-tier model cost a year earlier.

Two prices that moved since they were first set - and why that matters

o3 launched at $10 input / $40 output. It now runs at $2/$8 - a fifth of the original rate, following OpenAI's July 2026 price cuts. Claude Sonnet 5 was scheduled to rise from $2/$10 to $3/$15 on 1 September 2026; Anthropic cancelled that increase, so the launch price is still the current one.

Neither of those facts would show up in a snapshot pulled from an old announcement or a stale seed dataset - both numbers changed after launch, in opposite directions. It's a concrete reason to treat "current API pricing" as something to re-check at the point of use, not something to memorize from a launch-day headline.

What buying $10-per-million actually gets you now

A year or two ago, $10 per million output tokens was near the top of the market. Today it buys Gemini 2.5 Pro, GPT-4o or Claude Sonnet 5 - three models from three different labs, each with real production traction, each offering at least 128K tokens of context and two of the three offering a full 1M. The genuine premium tier - Claude Opus 5 at $25 output - is now a deliberate choice for specific, harder work, not the default entry point it once was.

The honest limit of this list

Eight models is not the whole board - it's the fraction of 88 that currently carries a verified, sourced price. Most of the board's models don't yet, and a blank price field means "not yet documented," not "free" or "unpriced by design." Treat this table as a snapshot of what's confirmed, not a comprehensive market survey - and re-check the number before committing a production budget to it, given how much movement just these eight prices have already shown.

---

Pricing confirmed live on 31 August 2026, sourced per model: Gemini 3.7 Flash, o3, Gemini 2.5 Pro, GPT-4o, Claude Sonnet 5, Qwen3.8-Max, Grok 4.6, Claude Opus 5.