GPT-5 Shuts Down on 10 December 2026
GPT-5, GPT-5-mini and GPT-5-nano lose API access on 10 December 2026. There is no like-for-like replacement at the same price inside OpenAI's current line, so every migration is also a price change.
The naming makes gpt-5.6-terra look like an incremental successor to GPT-5, but at $2.00/$12.00 per million tokens against GPT-5's $1.25/$10.00 it is 24% more expensive per million in plus a million out, by our arithmetic on the listed prices.
Below is every current price, the arithmetic on each swap, and where the cheapest replacement sits. AI Perks tracks $7.7M in credits across 194 companies for teams who would rather not pay the difference.
What You Pay Now and What the Replacements Cost
The three shutting-down models were priced with an 8x output multiple. Everything replacing them sits at 5x to 6.25x, which changes which workloads get hurt.
Legacy prices, all USD per million tokens, input / output:
| Model (shutting down) | Input | Output | 1M in + 1M out | Output multiple |
|---|---|---|---|---|
| gpt-5 | $1.25 | $10.00 | $11.25 | 8.0x |
| gpt-5-mini | $0.25 | $2.00 | $2.25 | 8.0x |
| gpt-5-nano | $0.05 | $0.40 | $0.45 | 8.0x |
OpenAI's current line, same units:
| Current model | Input | Output | 1M in + 1M out |
|---|---|---|---|
| gpt-6-astra (flagship) | $10.00 | $50.00 | $60.00 |
| gpt-6-astra above 200k context | $20.00 | $75.00 | $95.00 |
| gpt-5.5 | $5.00 | $30.00 | $35.00 |
| gpt-5.6-sol | $4.00 | $20.00 | $24.00 |
| gpt-5.4 | $2.50 | $15.00 | $17.50 |
| gpt-5.6-terra | $2.00 | $12.00 | $14.00 |
| gpt-5.4-mini | $0.75 | $4.50 | $5.25 |
| gpt-5.6-luna | $0.20 | $1.20 | $1.40 |
| gpt-5.4-nano | $0.20 | $1.25 | $1.45 |
The Sol, Terra and Luna names replaced the older Instant, Thinking, mini and nano scheme in July 2026. The combined column is our arithmetic on those listed prices, not an OpenAI figure.
Which Model Should Replace GPT-5, GPT-5-mini and GPT-5-nano?
Migrate GPT-5 to Claude Sonnet 5, GPT-5-mini to gpt-5.6-luna, and accept that GPT-5-nano has no cheap successor anywhere.
Estimated cost change per million in plus a million out, derived from the listed prices:
| If you run | Cheapest OpenAI successor | Change | Cheapest at that tier, any vendor | Change |
|---|---|---|---|---|
| gpt-5 ($11.25) | gpt-5.6-terra ($14.00) | +24.4% | Claude Sonnet 5 ($12.00) | +6.7% |
| gpt-5-mini ($2.25) | gpt-5.6-luna ($1.40) | -37.8% | gpt-5.6-luna ($1.40) | -37.8% |
| gpt-5-nano ($0.45) | gpt-5.6-luna ($1.40) | +211.1% | Meta Muse Spark Contributor ($0.30) | -33.3% |
Three things fall out of that table.
GPT-5-mini users get a discount. Moving to gpt-5.6-luna is roughly 38% cheaper than what they pay today. This is the only tier where the shutdown saves money.
GPT-5-nano users take the hit. The cheapest model in OpenAI's current line costs about 3.1x what GPT-5-nano costs. Meta's Muse Spark Contributor tier at $0.10/$0.20 is cheaper, but Meta trains on your prompts and completions there, which rules it out for most production traffic.
GPT-5 users should look outside OpenAI. That comes down to one Anthropic decision.
The Anthropic Price Cancellation That Changes the Answer
Anthropic had scheduled Claude Sonnet 5 to rise from $2/$10 to $3/$15 on 1 September 2026, then cancelled the increase. That cancellation is what makes Sonnet 5 the cheapest same-tier GPT-5 replacement.
Verbatim from Anthropic's pricing page: "The previously scheduled increase to $3/$15 per million input/output tokens on September 1, 2026 will not occur." The $2/$10 price, originally announced as introductory through 31 August 2026, is now standard.
At $3/$15, Sonnet 5 would cost $18.00 per million in plus out, 60% above GPT-5 and 29% above gpt-5.6-terra. The cancellation moved it to $12.00, 14% below terra. One reversed pricing decision flipped which vendor wins the GPT-5 migration, and almost nobody has written it up.
There is a sharper detail underneath. Claude Sonnet 5's output price is $10.00 per million, exactly what GPT-5 charged. Only the input side differs, $2.00 against $1.25. For output-heavy workloads, that swap is close to flat.
Anthropic's current line:
| Anthropic model | Input | Output | 1M in + 1M out |
|---|---|---|---|
| Claude Fable 5.1 (Mythos-class) | $10.00 | $50.00 | $60.00 |
| Claude Mythos 5.1 (limited availability) | $10.00 | $50.00 | $60.00 |
| Claude Opus 5 | $5.00 | $25.00 | $30.00 |
| Claude Sonnet 5 | $2.00 | $10.00 | $12.00 |
| Claude Haiku 4.5 | $1.00 | $5.00 | $6.00 |
Opus 5, launched 24 July 2026, holds the same $5/$25 price as Opus 4.8. Claude Haiku 4.5 at $6.00 combined is 47% cheaper than GPT-5, the move for any GPT-5 workload that can drop a capability tier.
What Migration Costs at Real Volume
Your input to output ratio decides the size of the bill increase, not the model you pick. The same migration ranges from +23% to +1.5% depending on the mix.
Estimated monthly cost at 50M input and 10M output tokens, an input-heavy mix typical of retrieval and long-prompt applications:
| Model | Monthly | vs GPT-5 | Annual difference |
|---|---|---|---|
| gpt-5 (shutting down) | $162.50 | baseline | - |
| Claude Haiku 4.5 | $100.00 | -38.5% | -$750 |
| Claude Sonnet 5 | $200.00 | +23.1% | +$450 |
| gpt-5.6-terra | $220.00 | +35.4% | +$690 |
| gpt-5.4 | $275.00 | +69.2% | +$1,350 |
| gpt-5.6-sol | $400.00 | +146.2% | +$2,850 |
| gpt-6-astra | $1,000.00 | +515.4% | +$10,050 |
Now flip the ratio to 10M input and 50M output, typical of generation and agent workloads:
| Model | Monthly | vs GPT-5 |
|---|---|---|
| gpt-5 (shutting down) | $512.50 | baseline |
| Claude Haiku 4.5 | $260.00 | -49.3% |
| Claude Sonnet 5 | $520.00 | +1.5% |
| gpt-5.6-terra | $620.00 | +21.0% |
| gpt-5.4 | $775.00 | +51.2% |
Sonnet 5 goes from +23.1% to +1.5% purely on workload shape, because its output price matches GPT-5's exactly. Measure your ratio before you pick. All figures above are estimates derived from the listed per-million prices, with no caching or batching assumptions applied.
Meta Muse Spark 1.1, 1.2 and 1.3 are listed at $1.25/$4.25, and Muse Image at $0.01 per image. Meta runs no startup credit program for Muse, so that bill is yours in full. AI Perks covers the providers that do.
What the Vendors Have Not Published
Several numbers quoted on other migration pages are not in any vendor's public documentation, and we will not invent them.
- Benchmark scores and capability rankings. No MMLU, HumanEval or SWE-bench figures for these models have been verified. Any guide ranking gpt-6-astra against Claude Fable 5.1 on a benchmark is guessing.
- Context window sizes. The one published threshold we can confirm is OpenAI's long-context tier for gpt-6-astra, which applies above 200k tokens at $20.00/$75.00. Everything else is unpublished.
- xAI and Grok token prices. Sources conflict badly enough that we will not print a number. Pricing varies by model, so check xAI's own console.
- Google Gemini prices. Not verified in this research, so Gemini is absent from every table above rather than guessed at.
- Rate limits and free-tier ceilings. Not published consistently by any provider here.
Every price here was checked on 16 September 2026 against OpenAI's and Anthropic's own pricing documentation. Most competing content still recycles pre-launch rumour.
How Free Credits Cover the Increase
The GPT-5 migration adds $450 to $10,050 a year at the volumes above. Provider credit programs cover that range comfortably, and most teams claim from one program when several apply.
$7.7M in tracked credits across 194 companies sits against a migration penalty that, for a typical mid-size application, runs a few hundred dollars a year.
Filter by provider. getaiperks.com covers the model providers you are migrating between. OpenAI and Anthropic both run credit programs; Meta does not.
Measure your input to output ratio first. The tables above show the same migration costing +23.1% or +1.5% on that variable alone. Pick the model after you know the number.
Claim before you cut over. Credits in place before migration absorb the higher rate from day one, not after the first surprising invoice.
Hold credits with more than one provider. The cheapest same-tier GPT-5 successor is at Anthropic, the cheapest GPT-5-mini successor at OpenAI. A single-vendor credit position forces the wrong choice on one of them.
Frequently Asked Questions
When exactly does GPT-5 stop working?
GPT-5, GPT-5-mini and GPT-5-nano lose API access on 10 December 2026. Do not start new builds on them. Existing integrations need a replacement model selected and tested before that date, because there is no same-price successor in OpenAI's current line.
What is the cheapest replacement for GPT-5?
Claude Sonnet 5 at $2/$10 per million tokens if you hold the tier steady: $12.00 per million in plus a million out against GPT-5's $11.25, and 14% below OpenAI's comparable tier, gpt-5.6-terra. Claude Haiku 4.5 at $6.00 combined is cheaper still if you can drop a tier. Credit programs for both providers are tracked at getaiperks.com.
Will my API bill go up after migrating?
Usually yes, by 6.7% to 24.4% per million tokens for GPT-5 workloads depending on the model you choose. GPT-5-mini workloads get about 38% cheaper on gpt-5.6-luna. GPT-5-nano workloads face roughly a 3.1x increase, which is the worst outcome of the three tiers.
Is gpt-6-astra worth the price for migration?
At $10/$50 per million tokens it costs 5.3x what GPT-5 cost, and above 200k context it rises to $20/$75. That is a flagship-tier decision, not a migration decision. Most GPT-5 traffic does not need it. Free credits make testing it cheap at getaiperks.com.
Did Claude Sonnet 5 get more expensive in September 2026?
No. Anthropic had scheduled an increase to $3/$15 per million on 1 September 2026 and cancelled it. The $2/$10 price, first announced as introductory through 31 August 2026, is now the standard price. That cancellation is why Sonnet 5 undercuts OpenAI's comparable tier.
How do I offset the higher per-token cost?
Provider credit programs cover the difference outright at typical volumes, where the annual increase runs from roughly $450 to a few thousand dollars. AI Perks tracks $7.7M in credits across 194 companies at getaiperks.com, covering both of the providers involved in this migration.
The models change every quarter. The bill does not have to.