Skip to content
Rush Commerce
AI & Automation3 min read

GPT-6.2 Astra cuts API price to $8/$40: rerun your routing

OpenAI's GPT-6.2 Astra drops to $8/$40 per million tokens with $0.40 cached input; GPT-6.2 Sol holds at $2/$10. Here is what to retest before you switch.

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, and the headline for anyone paying API bills is price. GPT-6.2 Astra lists at $8 per million input tokens and $40 per million output tokens in OpenAI's model catalog, down 20% from GPT-6 Astra's $10/$50. Sol stays at $2/$10. A cheaper flagship changes the math on which tasks you send where. It does not change it for you automatically.

What actually happened

From OpenAI's own API model docs:

  • GPT-6.2 Astra (gpt-6.2-astra): $8 input / $40 output per million tokens. OpenAI calls it the flagship and its recommended starting point.
  • GPT-6.2 Sol (gpt-6.2-sol): $2 input / $10 output per million tokens.
  • Both: 1.05M-token context window, 128K max output, reasoning effort from low to max, and support for function calling, web search, file search and computer use.

The bigger cut is on cached input. 9to5Google reports Astra's cached input at $0.40 per million tokens, and Artificial Analysis puts the previous rate at $1.00. That is a 60% cut on the part of the bill that long system prompts and agent loops live in. Sol's cached input is $0.10.

In ChatGPT, both models went to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, per 9to5Google. OpenAI's launch charts put the pair roughly level with Anthropic's Claude Fable 5.2 at lower cost. Those are vendor benchmarks; we have not seen independent numbers yet.

One safety note worth reading twice: according to Digital Trends, OpenAI says the new models are less likely to take improper actions, but its testing found they sometimes fail to tell users about problems they hit during a task.

Why it matters for your business

If you set up model routing on GPT-6 pricing, your thresholds are now stale. Tasks you pushed down to Sol to save money may now be cheap enough on Astra to justify the accuracy. And any workflow with a big, stable prompt prefix (an agent's instructions, a product catalog, a policy doc) gets a lot cheaper if your cache hits hold.

What we would do this week:

  • Rerun 20 to 50 real tasks on both 6.2 models. Track cost per completed task, not cost per token.
  • Check your cache hit rate. A 60% cached-input cut is worth nothing if your prefix changes every call.
  • Pin the model ID (gpt-6.2-astra) in production instead of an alias, so a default change does not ship to customers untested.
  • Log what your agents actually did, not just what they report. OpenAI's own note says the model may not tell you when something went wrong.

Key takeaways

  • GPT-6.2 Astra: $8/$40 per million tokens, down from $10/$50
  • Astra cached input reported at $0.40, down from $1.00; Sol cached input $0.10
  • GPT-6.2 Sol holds at $2/$10; both have a 1.05M context window
  • Benchmarks so far are OpenAI's own; test on your tasks before switching
  • OpenAI says the models can fail to report problems mid-task, so keep independent logs

Price changes should be a config change, not a rebuild. We build model routing with per-task cost tracking, so a new price list means updating a threshold, not rewriting your app. See how we build or run the numbers on your workflow.

Sources: OpenAI API model docs, 9to5Google, Digital Trends.

  • #gpt-6-2
  • #openai
  • #ai-costs
  • #model-routing
  • #prompt-caching
TR

Tommy Rush — Founder, Rush Commerce

Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More

Get The Rush Report weekly — one email, zero fluff.