AI pricing report: Aug 11–24, 2026

AI model prices move faster than almost any other software category, so we publish a short report each time we re-verify the catalog. This one covers the changes we tracked between 11 August and 24 August 2026 — every number below is the current verified figure, with the date it was checked.

The headline

The frontier got cheaper at the top and more crowded in the middle. Two flagship-grade models now cost a fraction of what GPT-5.6 Sol and Claude Opus 5 charge, DeepSeek’s value king went GA, and subscription pricing is fragmenting into more tiers.

The movers

Gemini 3.7 Flash launched (13 Aug) — and undercut its own line

Google’s newest Flash model arrived at $0.75 / $3.75 per 1M tokens with an intelligence index of 56 — cheaper and higher-indexed than Gemini 3.6 Flash, and with the same 1M-token context. For long-context work on a budget it’s now one of the best intelligence-per-dollar scores we track. Note that the current price is a promotional rate through 31 Dec 2026, after which it lists at $1.50 / $7.50.

DeepSeek V4 Pro went GA (13 Aug)

The 0813 snapshot took DeepSeek V4 Pro from a preview index of 45 to a GA index of 53, firmly frontier-grade. Its price now lists at $0.66 / $1.98 per 1M tokens (the official off-peak rate; peak hours are 2x) — still a fraction of the OpenAI and Anthropic flagships. It remains the reference “value frontier” model on this site.

Grok 4.6 (12 Aug) — flagship index, mid-tier price

xAI released Grok 4.6 at $2 / $6 per 1M tokens with an index of 61 — matching GPT-5.6 Sol’s index at roughly a tenth of its output price. For output-heavy workloads it’s the strongest value we’ve seen at the very top of the intelligence scale.

GPT-5.6 Sol’s official price pinned (24 Aug)

After our re-verification against OpenAI’s own pricing page, GPT-5.6 Sol is pinned at $4 / $20 per 1M tokens — the official list price, with a promotional rate through 21 Nov 2026. We now lock provider-sourced prices so the weekly OpenRouter refresh can’t drift them.

Two new frontier flagships (24 Aug)

• Claude Fable 5 — index 62, $10 / $50 per 1M. Anthropic’s premium adaptive-reasoning model and the current one-shot king. Expensive enough that it’s a subscription tool in practice.
• GLM-5.3 (max) — index 60, $1.40 / $4.40 per 1M. Z.ai’s DeepSeek-adjacent value play with genuinely strong front-end design, held back mainly by serving latency.

Subscription plans are splitting into more tiers

Three new plans appeared in the window: SuperGrok Lite ($10), SuperGrok Plus ($100) and Google AI Plus ($13.99). The pattern is clear — providers are pricing by workload depth, not just by “basic/premium”.

What to do about it

• Output-heavy, cost-sensitive work → Grok 4.6 is the current value leader at the frontier; DeepSeek V4 Pro and GLM-5.3 win if you want open weights.
• Long-context / budget → Gemini 3.7 Flash is hard to beat while the promo lasts.
• One-shot quality over price → Claude Fable 5, if your budget allows it.
• Re-check before you buy — every figure on this site carries its last-verified date, and the calculator recomputes value for your exact budget and workload.

Full change history is on the What’s new page.