GPT-5.6 builder's guide: same results, $33.27 down to $1.33
- OpenAI's new cost-cutting guide for developers shows a search benchmark score of 84, achieved for $33.27 with the previous top model and $1.33 with this generation's smallest model.
- Six production agent companies each report real metrics: 21% fewer input tokens, 28% less uncached input, and costs down to 1/18th—all from configuration changes, not a new model.
- The official charts also reveal something the text misses: turning reasoning effort all the way off makes the bill go up, not down.
OpenAI has published a GPT-5.6 builder's guide focused on cutting agent costs: which model to choose, how much reasoning effort to set, and how to use three new API switches—all direct configuration changes. Six startups already running agents in production each contributed real-world metrics.
The guide challenges the default way agents have been built: reach for the most expensive flagship model and crank up reasoning. That became the default because only top-tier models could handle long contexts and heavy tool use. With the 5.6 generation, smaller models can think a little longer and match the previous flagship's performance at a fraction of the price.
A roadmap: six levers you can pull today
Saving money works on two levels. The first level is two dials—reasoning effort and model choice—which are simple configuration changes. The second level involves three architectural switches and a caching breakpoint, which change how requests are structured. The table below is your roadmap: the first two rows cover the dials, the last four cover the switches, and each gets its own dedicated section later in this piece.
The first dial: a lower effort level beats last gen's highest
Since GPT-5, each generation has chased the same goal: accomplish more with fewer tokens. GPT-5.6 continues this trend—the Harness (the orchestration layer around the model: how requests are sent, tools are called, and loops are run) is unchanged, yet performance climbs and costs fall.
The guide's most direct example: with the Harness completely fixed, GPT-5.6 Sol at 'Low' reasoning effort outscores GPT-5.5 at 'High' on the Agents' Last Exam benchmark. Since effort is the main cost lever, winning this matchup means the same task just got a whole price tier cheaper. The guide doesn't publish specific scores for this particular comparison.
For a direct comparison you can verify yourself, here are three matchups, all drawn from real data points on the benchmark charts. Old model on the left, new model on the right:
The cost scales on the two charts are very different: the search benchmark costs a few dollars to run completely, while the coding index costs hundreds or thousands—the topics and task complexity aren't comparable. Don't compare dollar amounts across the two charts. Each comparison above is within the same chart.
So what are you actually saving on? It's the wasted effort. Hex, a data analytics company, dropped the model into their existing Harness and found the lowest effort setting gave them the best results they'd ever seen:
We put GPT-5.6 into our existing Harness, and the low reasoning setting gave us the best results we've had. It knows when the data just isn't there, doesn't chase dead ends, and gets to the right answer with fewer tokens.
Izzy Miller, Head of AI Research at Hex
That phrase—"knows when the data isn't there"—is the key. Older models would rephrase a search query and keep digging when they couldn't find something, paying for every tool result that came back as input tokens. A model that knows when to stop saves exactly those wasted cycles.
Lowering Effort Works, But Turning It Off Completely Costs More
Lowering the effort level saves money. So wouldn't turning it off entirely save even more? Looking at the actual data points on OpenAI's charts, the answer is no.
The dial below lets you explore this: pick a reasoning level, and watch the cost and score for all four models change. The cost bars are scaled so GPT-5.5 at Xhigh equals the full length of the track ($33.27 on the Search benchmark).
Now move the dial all the way to the left. You'll see something counterintuitive: turning reasoning off does not save money. There's evidence for this on both official charts:
- Sol on the Search benchmark: With reasoning off, it costs $1.97 for a score of 69.67. At the lowest effort setting, it costs just $0.50 for a score of 65.32. So turning reasoning off didn't save money—it cost 3.9x more for only 4.3 extra points.
- Luna on the Coding Index: With reasoning off, it costs $123.66 for a score of 37.30. At the lowest effort setting, it costs $74.84 for a score of 42.40. This one is both more expensive and worse—65% more money for 5.1 fewer points.
There's a similar inflection point at the other end. On Search, Luna going from "Xhigh" to "Max" sees its bill rise from $1.33 to $1.42, but its score drops from 84.04 to 83.25. Maxing out the dial doesn't necessarily help either.
If the model isn't allowed to think, it has to stumble toward an answer by making more tool calls and going around in circles. And every tool result that comes back is billed as an input token. Thinking costs money, but it buys "fewer wrong turns"—a deal that usually pays off on hard tasks.
The right way to tune effort is to start from the bottom and work your way up, watching cost and score move together until you hit the sweet spot—not to start from the top and cut downward. The lowest effort setting is usually the best value. The far-left setting, completely off, doesn't pay off on either of these charts.
The second dial: smaller models can save 25x
Once you've adjusted the effort dial, the next one is model choice. GPT-5.6 comes in three sizes: Sol (flagship), Terra (mid-size), and Luna (small).
In the past, the smaller models were only good for edge cases—they couldn't handle long contexts or heavy tool use, and they'd fall apart once the task chain got long. That's changed with this generation, thanks to test-time compute.
This lets the model think a bit longer before answering. The thinking time is compute, and compute is money. It's the difference between giving a student an extra ten minutes to check their work versus swapping in a smarter student. The smaller models take the first path; the model itself doesn't change.
The result is that Luna and Terra can often match GPT-5.4 or GPT-5.5 performance at a tiny fraction of the cost. The guide's headline comparison comes from BrowseComp, a benchmark that tests the ability to unearth obscure facts: