The same model is published at $2.00 to $16.00 per million input tokens and $10.00 to $60.00 per million output tokens depending on the meter, before data residency adds a documented uplift. The $4.00 / $20.00 deck most teams budget against is promotional, and the registry’s rate-card source still carries $5.00 / $30.00 for the gpt-5.6 alias.
per 1M tokens, short context — promotional
$2.00 → $16.00 input, Global meters
Nov 21, 2026 — “at least through”
USD per one million tokens, standard tier unless stated. OpenAI rates observed September 21, 2026; Azure rates from Microsoft’s Retail Prices API for the Standard Global meters.
| Meter | Where it is published | Input | Output | Note |
|---|---|---|---|---|
| Standard · short context (≤272K) | OpenAI direct · Azure Global | $4.00 | $20.00 | The promotional deck, and the one the registry records. Azure meter: 5.6 sol ShortCo Std Gl. |
| Standard · long context | Azure Global · OpenAI direct | $8.00 | $30.00 | Azure prices long context as a separate meter, not a multiplier. |
| Batch | OpenAI direct | $2.00 | $10.00 | −50% on both legs vs standard. |
| Flex | OpenAI direct | $2.00 | $10.00 | −50% on both legs vs standard. |
| Fast mode (Azure: Priority Processing) | OpenAI direct · Azure Global | $8.00 | $40.00 | 2× input and 2× output. Renamed from Priority Processing on Jul 30, 2026. |
| Fast mode · long context | Azure Global | $16.00 | $60.00 | 4× the standard short-context input rate. |
Sources: developers.openai.com · API pricing (decks, Fast mode, batch, flex, promotional note) · prices.azure.com · Azure Retail Prices API (Azure meters) · developers.openai.com · API changelog (Fast mode rename, Jul 30, 2026)
OpenAI announced the cut on August 21, 2026 and attached a qualifier rather than an end date.
On August 21, 2026 OpenAI stated that GPT-5.6 Sol “now costs $4 per million input tokens and $20 per million output tokens, representing 20% lower input pricing and 33% lower output pricing”, and that the promotional pricing “is available at least through November 21, 2026”. The current pricing page repeats the same wording.
A −20% input cut and a −33% output cut resolve to a pre-promotional deck of $5.00 input / $30.00 output ($4.00 ÷ 0.80 and $20.00 ÷ 0.667). That is exactly the deck still carried today by the Azure gpt-5.6 alias record, the Perplexity record, and gpt-5.5. No official page states what the deck becomes after November 21 — “at least through” is a floor, not a guarantee.
A November-and-beyond forecast built on $20.00 output should carry the $30.00 case. At the current promotional deck the difference is a 50% output increase on every uncached output token — larger than most routing optimisations recover.
ALLM registry records for the GPT-5.6 Sol family, observed September 20, 2026, compared against the official decks above.
| Registry record | Provider | Region | Input | Output | Against the official deck |
|---|---|---|---|---|---|
| gpt-5.6-sol | OpenAI direct | Global | $4.00 | $20.00 | Matches the official standard deck |
| gpt-5.6-sol | Azure AI | Global | $4.00 | $20.00 | Matches Azure’s own Standard Global meter |
| global.openai.gpt-5.6-sol | AWS Bedrock | Global | $4.00 | $20.00 | Matches OpenAI direct |
| eu/gpt-5.6-sol | Azure AI | EU Data Zone | $4.40 | $22.00 | Exactly +10% |
| us/gpt-5.6-sol | Azure AI | US Data Zone | $4.40 | $22.00 | Exactly +10% |
| us.openai.gpt-5.6-sol | AWS Bedrock | US | $4.40 | $22.00 | Exactly +10% |
| openai/gpt-5.6-sol | OpenRouter | Global | $2.00 | $10.00 | Matches OpenAI’s Batch / Flex tier rate |
| openai/gpt-5.6-sol | Perplexity | Global | $5.00 | $30.00 | Pre-promotional deck |
| gpt-5.6 (alias) | Azure AI | Global | $5.00 | $30.00 | Pre-promotional deck |
| eu/gpt-5.6 · us/gpt-5.6 | Azure AI | EU / US Data Zone | $5.50 | $33.00 | Pre-promotional deck + 10% |
Registry rate cards for these deployments cite litellm · model_prices_and_context_window.json as their source URL. Compare with prices.azure.com · Azure Retail Prices API and developers.openai.com · API pricing.
OpenAI states that “the gpt-5.6 alias routes requests to gpt-5.6-sol”. The registry prices those two names differently — on the same provider, in the same region.
The registry carries azure-ai · gpt-5.6-sol at $4.00 / $20.00 and azure-ai · gpt-5.6 at $5.00 / $30.00 — a 25% higher output rate for a name that resolves to the same model. On OpenAI direct both names carry $4.00 / $20.00, so the split is specific to how the Azure records were sourced.
Querying Microsoft’s Retail Prices API for the Foundry Models service returns 96 distinct meters containing “5.6”, and every one of them is 5.6 sol, 5.6 terra or 5.6 luna. There is no separately priced bare 5.6 SKU — the $5.00 / $30.00 alias record does not correspond to a published Azure meter.
Every non-global record in the table above lands at exactly +10% — Azure EU, Azure US and AWS Bedrock US. That matches OpenAI’s published rule that regional processing (data residency) endpoints “are charged a 10% uplift for models released on or after March 5, 2026”. Azure sells the same model under three deployment shapes — Global Deployment (Global SKU), Data Zone Deployment (geographic, EU or US) and Regional Deployment (local region, up to 27 regions) — which is why the registry carries eu/ and us/ prefixed records alongside global ones. Azure’s own Data Zone meters price 5.6 sol ShortCo at $4.40 / $22.00 against $4.00 / $20.00 Global. Residency is a price line, not a footnote.
Officially announced changes that alter a migration deadline, a retry path, or a provisioning control.
Announced September 11, 2026: gpt-5.4-cyber “will be removed from the API on October 1, 2026”, with gpt-5.6-cyber as the recommended replacement. Note the price delta: gpt-5.4-cyber is not listed on the current pricing page at all, while gpt-5.6-cyber is $12.50 / $75.00. Migrating is a budget change, not only an endpoint change.
Since September 2, 2026, traffic that increases too quickly returns 429 with the slow_down code, while temporary model overload returns 503 with server_is_overloaded. Identical payloads, opposite remedies: the first is a rate-limit response that a faster retry makes worse, the second is an availability response that a fallback route should absorb. A retry policy that treats both as “retry and back off” will amplify its own throttling.
September 10 gave project API keys an expiry date and let administrators cap key lifetime at organization or project level, so newly created keys must expire within the configured limit. September 15 added creation governance: allow only service-account keys, allow only user-owned project keys, or disable key creation entirely, with organization restrictions taking precedence. Existing keys are unaffected — which is the gap to close, since the control does not retroactively shorten the credentials already in your deployments.
Official pricing pages and the Azure Retail Prices API observed September 21, 2026 (Europe/Paris). ALLM registry records observed September 20, 2026.
Related live views: the OpenAI vs Azure AI deployment comparison and the non-token meter report render the current catalog records.
Deployment intelligence, explained
A model name does not determine what a token costs. For GPT-5.6 Sol the published Global meter deck spans $2.00 to $16.00 per million input tokens and $10.00 to $60.00 per million output tokens depending on the meter — standard short or long context, batch, flex, or fast/priority processing — and data residency adds a further documented uplift on top.
A registry rate card records one input/output pair per deployment, so it can only show the tier it sampled. Read a rate card as the floor of a route, confirm the tier your workload actually bills on, and treat any pricing date with the qualifier the provider attached to it.
We use technologies necessary for the website to operate. With your consent, we use PostHog and Google Analytics to measure pseudonymised audience data. Refusal does not affect access to the website.