AI Tools
GPT-6.2 pricing: what Astra and Sol cost per million tokens
GPT-6.2 pricing starts at $8 per million input tokens and $40 per million output tokens for Astra, 20% below GPT-6's $10 and $50. Sol runs $2 and $10, unchanged from GPT-6.1 Sol. Both new models ship with a 1.05M-token context window and 128K max output tokens.
What matters
- GPT-6.2 Astra lists at $8 and $40 per million input and output tokens, against $10 and $50 for GPT-6 Astra.
- Sol stays at $2 and $10 per million, identical to GPT-6.1 Sol, so swapping the model ID costs nothing in tokens.
- Cached input on Astra drops from $1.00 to $0.40 per million, the cut that matters for agents replaying long prompts.
- Ultrafast bills 3x standard until February 28, 2027, then doubles to 6x from March 1, 2027.
- Claude Fable 5.2 matches Astra at $8/$40 but charges $0.20 per million for cached reads, half of OpenAI's rate.
GPT-6.2 pricing by tier
OpenAI added two model IDs on November 17, 2026: gpt-6.2-astra and gpt-6.2-sol. GPT-6.2 Astra is the flagship, positioned for demanding reasoning and coding, with five reasoning settings from low to max. Sol is the everyday model OpenAI points agent and coding workloads at. Both accept text and images, return text, and hold a 1.05M-token context window with 128K of maximum output.
Standard rates, per million tokens:
- GPT-6.2 Astra: $8 input, $40 output, $0.40 cached input
- GPT-6.2 Sol: $2 input, $10 output, $0.10 cached input
- Cache writes: $10 per million on the 30-minute cache and $20 on the 12-hour cache for Astra; $2.50 and $5 for Sol
- Batch: half the standard rate, so $4/$20 for Astra and $1/$5 for Sol
- Fast: double standard, $16/$80 for Astra and $4/$20 for Sol
Prompt caching is priced in two parts on the same page, a read rate and a write rate, and the write rate depends on how long you want the cache held. That detail is where most quick cost estimates go wrong: a 12-hour cache write on Astra costs twice a 30-minute one.
Astra is 20% cheaper than the model it replaces
GPT-6 Astra listed at $10 input and $50 output in September. The new Astra cuts both to $8 and $40. Cached reads fall harder, from $1.00 to $0.40 per million, a 60% reduction, and the 30-minute cache write drops from $12.50 to $10.
The cache cut is the one that shows up on real invoices. A support agent that replays a 10,000-token instruction set 20,000 times a month pushes 200 million cached tokens. At the old $1.00 rate that is $200 of cache reads; at $0.40 it is $80. The output tokens are identical in that scenario, which is exactly why per-token comparisons that only quote input and output prices miss the shift.
Sol holds the line at $2 per million input tokens
GPT-6.2 Sol costs precisely what GPT-6.1 Sol cost: $2 input, $10 output, $0.10 cached input. OpenAI did not raise the small-model price to fund the flagship cut, so a team already running 6.1 Sol can move to the new ID without revising its budget. The knowledge cutoff moves to June 1, 2026 on both 6.2 models, against May 18, 2026 on GPT-6 Astra and GPT-6 Luna.
The cheaper rungs did not get a 6.2 refresh. GPT-6.1 Terra stays at $1/$5 and GPT-6 Luna at $0.10/$0.50, and both still appear on the pricing page, which means the practical ladder now runs Luna, Terra, Sol, Astra with a 4x step between Sol and Astra.
What a month of GPT-6.2 actually costs
Model a mid-size production workload: 2 million fresh input tokens, 8 million cached reads and 2 million output tokens per month, priced at standard rates.
- GPT-6.2 Astra: $99.20
- GPT-6 Astra: $128.00
- Claude Fable 5.2: $97.60
- GPT-6.2 Sol: $24.80
Moving that workload from GPT-6 Astra to 6.2 Astra saves $28.80 a month, 22.5%. Batch processing halves the Astra bill to $49.60 for the same tokens. Background requests take a different discount: 30% below standard on a two-hour completion window and 40% below on an eight-hour window, which lands the same workload at $69.44 and $59.52. Neither discount stacks with the other on the pages checked, so pick one path rather than assuming both apply.
The Ultrafast cliff lands on March 1, 2027
Ultrafast is the tier to watch before you benchmark latency against a budget. For GPT-6.2 Astra and Sol it costs 3 times standard through February 28, 2027: $24 input and $120 output per million on Astra, $6 and $30 on Sol.
From March 1, 2027 the multiplier becomes 6 times, and OpenAI already publishes the resulting list rates of $48 and $240 for Astra. That is a 2x step on the same model, on a date roughly three months out. If a latency-sensitive product is priced on today's Ultrafast rate, the margin math needs rework before Q1 ends.
Flex processing is the other gap. The Flex table lists gpt-6-astra, gpt-6.1-sol, gpt-6.1-terra and gpt-6-luna at half price, but neither 6.2 model, so the cheap tier is not available for the new IDs as the page stands today.
GPT-6.2 vs Claude: same headline, different cache math
Claude Fable 5.2 also lists $8 input and $40 output per million tokens. Claude Sonnet 5.5 matches Sol at $2 and $10. On standard list price the two flagship families are now identical, which is a change from September, when GPT-6 Astra at $10/$50 sat above Anthropic's flagship.
Cached reads break the tie. Anthropic charges $0.20 per million cached input tokens on Fable 5.2, half of OpenAI's $0.40 on Astra. Cache writes land in the same place: $10 per million on Fable 5.2, $10 on Astra for a 30-minute hold. Both vendors discount batch work by 50%.
So the decision rule is narrow. For chat-style traffic with little prompt reuse, the two flagships cost the same and the choice belongs to evals. For agent loops that resend a large context on every turn, Anthropic's read rate is worth real money at high volume, and OpenAI's answer is the 40% cheaper cache read than its own previous flagship rather than a lead over Claude.
Which GPT-6.2 model should you pick?
- Astra for hard reasoning, long-horizon coding and anything where a wrong answer costs more than the tokens.
- Sol for agents that run all day, classification at volume and first-pass code review.
- Luna at $0.10/$0.50 when the task is narrow and volume is the constraint.
One integration caveat before the swap: Chat Completions still works, but tool calling requires the Responses API. OpenAI also documents two behavioral changes in the GPT-6.2 migration guide, covering how reasoning is preserved when a conversation prefix is edited and how the top-level instructions field is handled. Both touch code paths that quietly shape output quality, so read the guide before changing the model ID in production.
At a glance
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Cached input (per 1M tokens) |
|---|---|---|---|
| GPT-6.2 Astra | $8.00 | $40.00 | $0.40 |
| GPT-6.2 Sol | $2.00 | $10.00 | $0.10 |
| GPT-6 Astra | $10.00 | $50.00 | $1.00 |
| GPT-6.1 Sol | $2.00 | $10.00 | $0.10 |
| Claude Fable 5.2 | $8.00 | $40.00 | $0.20 |
| Claude Sonnet 5.5 | $2.00 | $10.00 | $0.10 |
FAQ
Is GPT-6.2 cheaper than GPT-6?
Yes for Astra: $8 input and $40 output per million tokens against $10 and $50, a 20% cut, with cached reads down from $1.00 to $0.40. GPT-6.2 Sol costs the same as GPT-6.1 Sol at $2/$10, so the small model did not move.
What is the difference between GPT-6.2 Astra and Sol?
Astra is the flagship for demanding reasoning and coding, with reasoning effort settings from low to max. Sol is the cheaper everyday model for coding and agent workloads. The price gap is 4x on both input and output.
Does GPT-6.2 support batch pricing and prompt caching?
Both. Batch runs at half the standard rate, $4/$20 per million for Astra and $1/$5 for Sol. Prompt caching is available with $0.40 reads for Astra and writes at $10 for a 30-minute cache or $20 for a 12-hour cache.
Related reading
GPT-6 Astra pricing at launch, OpenAI API pricing overview, Claude Fable 5.2 pricing