How much does GPT-5.6 Sol cost now?

As of August 21, 2026, GPT-5.6 Sol costs $4 per million input tokens and $20 per million output tokens for standard short-context use, down from $5 and $30. That is a 20% cut on input and a 33% cut on output. On a typical 30,000-input / 5,000-output task the cost falls from about $0.30 to about $0.22.

Is the GPT-5.6 Sol price cut permanent?

It has been reported as a promotional cut running roughly three months rather than a permanent reprice, though OpenAI's pricing table now shows the lower rates. Plan capacity at the promotional rate but budget at the old rate. Teams that rebuilt unit economics around temporary pricing in 2026 — DeepSeek's August 16 reprice being the cautionary example — discovered the hard way that introductory rates revert.

Does this make GPT-5.6 Sol cheaper than Claude Opus 5?

Yes. Claude Opus 5 is $5 input and $25 output per million tokens, so Sol at $4/$20 is now cheaper on both sides. On a 30K-in/5K-out task Sol costs about $0.22 against roughly $0.275 for Opus 5, about 20% less. Before the cut Sol was slightly more expensive than Opus 5, so the ordering has genuinely flipped.

Why did OpenAI cut prices on its flagship model?

Competitive pressure at the frontier tier. Anthropic cancelled its scheduled Claude Sonnet 5 price increase in early August 2026, Google shipped Gemini 3.7 Flash at an introductory $0.75/$3.75 on August 13, and Grok 4.6 holds at $2/$6. Cutting the flagship rather than the cheap tier signals OpenAI is defending high-value agentic workloads, where token volumes are largest and switching costs are highest.

Quick Answer

OpenAI Cut GPT-5.6 Sol to $4/$20: What Changed

Published: August 25, 2026

The Short Answer

On August 21, 2026, OpenAI cut GPT-5.6 Sol from $5/$30 to $4/$20 per million input/output tokens — a 20% input cut and a 33% output cut. It is reported as a promotional window of roughly three months.

The effect: OpenAI’s flagship now costs less than Claude Opus 5, reversing the previous order.

The New Numbers

Model	Input / MTok	Output / MTok	30K-in / 5K-out task
GPT-5.6 Sol (new)	$4.00	$20.00	~$0.22
GPT-5.6 Sol (old)	$5.00	$30.00	~$0.30
Claude Opus 5	$5.00	$25.00	~$0.275
GPT-5.6 Terra	$2.00	$12.00	~$0.12
GPT-5.6 Luna	$0.20	$1.20	~$0.012
Grok 4.6	$2.00	$6.00	~$0.09
Gemini 3.7 Flash	$0.75	$3.75	~$0.041

Verified August 25, 2026. Sol Ultra remains a separate, higher tier. Rates apply to standard short-context use.

Why the Output Cut Matters More

The input cut is 20%; the output cut is 33%. That asymmetry is the substance of the announcement.

Agentic workloads are output-heavy. A coding agent that reads a repository once and then writes, revises and re-writes across dozens of turns generates far more output tokens relative to input than a summarization or classification job. Cutting output pricing by a third targets exactly the workload class where token spend is growing fastest and where Anthropic has held the strongest position.

Put differently: this is not a broad consumer price cut. It is a bid for agent traffic.

The Third Price Move This Quarter

Sol’s cut lands in a dense sequence:

July 30, 2026 — OpenAI cut Terra 20% and Luna 80%
August 10, 2026 (approx.) — Anthropic cancelled the scheduled September 1 rise of Claude Sonnet 5 to $3/$15; $2/$10 became permanent
August 13, 2026 — Google shipped Gemini 3.7 Flash at an introductory $0.75/$3.75, half its post-2026 rate
August 16, 2026 — DeepSeek raised V4 Pro and V4 Flash prices and introduced peak/off-peak splitting
August 21, 2026 — OpenAI cut Sol to $4/$20

The direction is not uniform. Western frontier labs are cutting to defend share while DeepSeek repriced upward now that it has volume. Anyone writing “AI inference is getting cheaper” as a trend line is describing half the market.

Should You Migrate?

If you already run on Sol: nothing to do. You get the lower rate automatically. Update your cost model and re-check whether workloads you pushed down to Terra for budget reasons now belong back on Sol.

If you run on Claude Opus 5 purely for cost: you were not, because Opus 5 was cheaper before. Now it is not, and a migration is worth evaluating — but not on price alone. Opus 5 retains an edge on long-horizon agentic coding, and its 1M context with 128K max output is a different shape from Sol’s. Run your own evaluation set before moving.

If price is your binding constraint: neither. Gemini 3.7 Flash at ~$0.041 per task is 19% of a Sol call, and Grok 4.6 at ~$0.09 is 41%. The frontier tier is the wrong tier for cost-sensitive work regardless of who is winning the frontier price war this month.

The Promotional Trap

The rates are live and published, but reported as promotional for roughly three months. The failure mode is predictable:

Team rebuilds unit economics at $4/$20
Team ships pricing to customers based on that margin
Promotion expires, rates return to $5/$30
Output-heavy workloads see a 50% cost increase on the output side

Budget at the pre-promotion rate and treat the difference as margin, not as capacity. If the promotional pricing becomes permanent, you have upside. If it expires, nothing breaks. DeepSeek customers who built on the flat $0.435/$0.87 V4 Pro rate before the August 16 reprice learned this in the same month.