共有:

Price War

Sam Altman Says It:
OpenAI Is Ready to Slash Prices

Through June it was a rumor cycle; on July 16, the OpenAI CEO said it himself. Targets: Anthropic and China. The shift from "premium first" to "volume matters" rewrites the cost model for anyone building on metered LLM APIs.

AI Navigate Editorial·2026.07.16·6 min read

$/1M tok Time OpenAI Anthropic DeepSeek 2024 H1'25 H2'25 H1'26 H2'26 2027 est. FRONTIER LLM API PRICE
FIG. If OpenAI opens the throttle, the three price curves converge toward the bottom.
01

The News

From rumor to on-the-record

On July 16, 2026, Sam Altman said the pricing strategy is being reset — publicly.

Altman said OpenAI is prepared to take a more aggressive pricing posture against Anthropic and the Chinese vendors — moving the story from "trade press speculation" to the CEO's own words. Details will land on OpenAI's official site and in the transcripts from Bloomberg.

The backdrop is clear. DeepSeek's $71B valuation and 3.5× revenue growth; Anthropic's extended free Fable 5 access. In a market where price gaps move customers, OpenAI has chosen to lean into volume. How Anthropic and DeepSeek respond next week will set the pace.

02

By the Numbers

What "serious price cut" really means

−40%
Market-expected size of next cut
$0.60
Expected GPT-5.6 mini per 1M output tok
3 vendors
Racing to the bottom
Q4 2026
Likely first repricing window

These figures are consensus estimates, not confirmed. Altman didn't quote a specific number. Given OpenAI's 12- to 24-month repricing pattern, expect roughly 30–50% off at the frontier tier and 40–60% off at mini/nano.


03

Why It Matters

Why this hits now

If your product is billed metered against LLM APIs, this lands directly on your P&L.

01

Cost basis shifts

A feature you'd designed at "$0.02 per call" suddenly costs 60% of that. Margins jump. Simultaneously, features that didn't clear the payback threshold now do — worth revisiting your backlog.

02

Pivot to volume

OpenAI shifts from premium-first to a volume+margin balance. With GPT-5.6 Sol's repeated safety incidents making "frontier only" untenable, the strategy is to make "good enough for everyone" the wedge.

03

Will the others follow?

Anthropic already responded via extended free Fable 5 access — but through "more usage at the same price", not a headline cut. DeepSeek, sitting on $71B, may prioritize growth investment over price cuts. It's not a given that all three race down at once.

04

Who's Affected

Who feels it, and how

Product engineers

Costs for metered API features move meaningfully within 6 months. Now is a fine time to plan caching layers and mini/nano-tier substitutions in your roadmap.

Executives & Finance

You can budget assuming AI unit costs fall year over year. The catch: confidence in a real cut is ~70–80%. Hold annual budgets against current pricing, but keep an activation plan for when the cut lands.

PMs & Product Owners

Line up features that need lower unit costs to work — long-form generation, live voice, agent chains. When the cut hits, launch immediately.

05

The Counterpoint

Downsides you shouldn't ignore

Cheap isn't always good.

①Fast cuts can thin SLA. GPT-5.6 Sol's alerts may be the leading edge of quality/safety strain. ②Lower prices can mean lower R&D total. Next-generation model progress could slow — you may pay in long-run quality for short-run cost. ③Compression squeezes thin-wrapper startups. SaaS built as bare API wrappers, without proprietary data or domain differentiation, will find 2027 harder than 2026.


06

What to Do Next

Recommended actions

Short term (0-3 months)Medium term (3-12 months)
Break out GPT/Claude/DeepSeek call costs by featurePrepare a "post-cut" feature campaign as a PoC
Identify caching + mini/nano substitution candidatesScore vendors on SLA, audit log, and data residency again
Hold annual budgets at current pricing; earmark upsideIf you rely on thin-wrapper SaaS, plan replacement paths

Altman's statement is a signal that AI unit costs still have one or two more steps down. Hobbyist and personal-use bills won't change much — but enterprise cost assumptions will move visibly within six months. The right stance is not just to welcome the cut, but to prepare for it. That's the operational summary.