6. Thinking Longer

Page 5 gave the model a chain. A chain has a length, and that length is compute you pay for, one token at a time. So put the length on a dial and turn it.

1. Cut it off, make it answerone problem · one model · the cap on a dial

The dial below is max_new_tokens, a hard limit on how much the model may write. Turn it down and the chain stops mid-thought, so at the cap we close its working off, start the answer line for it, and allow it 32 more tokens to finish.reference 1 That is why a starved model still produces an answer to grade rather than nothing at all. The tinted band in the trace below is that line, printed as we wrote it. Every figure on this page carries its own copy of this dial, so nothing you do here moves anything below.

Qwen2.5-3B-Instruct at bf16, greedy · MATH-500 level 4, Counting & Probability

Suppose that I have 6 different books, 2 of which are math books. In how many ways can I stack my 6 books on a shelf if I do not want the math books to be next to each other?

Token budget

This chain runs to 351 tokens and stops on its own, so the dial bites at 128 and 256 and then has nothing left to cut: 512 and up are the same chain.

the model →

To solve this problem, we need to calculate the number of ways to arrange 6 books such that the 2 math books are not next to each other. 1. Calculate the total number of arrangements without any restrictions: The total number of ways to arrange 6 books is given by the factorial of 6, denoted as 6!: 6! = 720 2. Calculate the number of arrangements where the two math books are together: - Treat the two math books as a single "super book". This reduces the problem to arranging 5items (the "super book" and the other 4 books). - The number of ways to arrange these 5 items is 5!: 5! = 120 - Within the "super book", the two math books can be arranged in 2! ways: 2! = 2 - Therefore, the total number of arrangements where the two math books are together is: 5! × 2! = 120 × 2 = 240 3. Calculate the number of valid arrangements where the two math books are not next to each other: Subtract the number of arrangements where the math books are together from the total number of arrangements: 6! - (5! × 2!) = 720 - 240 = 480 The final answer is: 480

budget spent · we close the block and make it commit

Final answer: 360

✗ graded wrong128 of 128 tokens thoughtanswered 360 · the answer is 480

or run a different problem · 8 of these ship with their full traces

The grey text is what the model would have written and was not allowed to; the tinted line is ours, not its conclusion. Step through the stops and the verdict does not slide, it flips: for one problem, thinking longer is a threshold and not a gradient. That is Act 1's slow deliberate System 2, priced in the only currency a model has.

2. The same dial, every problemthree checkpoints · the same 100 problems

One problem is an anecdote. Below is the same sweep run over all 100, on three checkpoints: how many each solves at every stop on the dial. The line on the chart is this figure's own dial, so drag it. Read the shape rather than the heights.

past here, doubling stops paying for the 3B0204060128256512102420483B1.5B7B128token budget (max_new_tokens), log scalesolved (%)

at 128 tokens, 3B-Instruct solves

7 of 100

tokens generated to get there

13,211

the cap bound 100% of these 100 chains

the procedure · Cap generation at N tokens; if the model is still inside its think block, close it and append "Final answer: \boxed{"; allow 32 more tokens; grade with MATH-500's reference answer. Repeat for N in {128, 256, 512, 1024, 2048} over 100 problems from levels 4-5, greedy throughout.

Every curve climbs steeply and then enters the same slab: past 1024 tokens, one more doubling moves the 3B down by 1 point, which is one problem out of 100. Three heights, one shape, and the shape is the part that is not about any of these checkpoints.

3. It stops mattering before you run out of dialone square per problem · placed by how long its chain ran

So why does the curve flatten? Below are the same 100 problems again, each one placed by how long its chain ran when nothing cut it off. The line is the budget you are holding: everything to its left finished on its own, everything to its right was still being cut, and here the line is the control: drag it along the axis and watch it move past the pile.

051210241536204820480 finished on their own100 still being cut128tokens the chain actually ran to, with nothing cutting it
solved at some budget (49) never solved (51) dashed and off the axis: still running at 2048, so its length is a floor (6) the budget you are holding · drag it

and where each problem first grades correct

1287
25611
51220
102411
20480
never51

At 2048 tokens the shaded strip holds 6 squares. Every other chain stopped when it had nothing left to say, a median of 655 tokens for the ones that never come right, with 1393 still unspent. So 46 of these problems did not run out of room, they ran out of ideas. The dial was never what was holding them back.

4. The other direction: refuse to let it stopthe same 100 problems · not allowed to stop

Everything above takes room away. This does the opposite: when the model tries to end its turn, the stop is dropped, the word Wait is appended to what it has written, and it carries on. 3 times over.

allowed to stop0 injectionsstop suppressed3 injections43 right57 wrong42 right58 wrong10 right → wrong9 wrong → right
wrong, then right once it could not stop (9) right, then wrong (10) right either way (33) wrong either way (48)

9 answers turned from wrong to right against 10 that turned from right to wrong, p = 1.000. Both crossings are thick and they cancel. In all, 30% of answers moved, because a further 11 swapped one wrong answer for another and never touched the verdict.

and the same crossing on all three checkpoints

Answers turned right and turned wrong by 3 injections, one row per checkpoint, against what each solved when allowed to stop.
checkpointright → wrongwrong → rightsolved, and what it cost
3B10
943%42% · 30% of answers moved · ×1.8 the tokens · p = 1.000
1.5B9
722%20% · 59% of answers moved · ×2.2 the tokens · p = 0.804
7B7
460%57% · 21% of answers moved · ×1.6 the tokens · p = 0.549

Both counts are against the same problem at zero injections, so they are paired.

It redistributes rather than adds. Spending 1.6 to 2.2 times the tokens buys a third of the answers changed and no more of them right: across the three checkpoints it turns 20 wrong answers right and 26 right answers wrong. Not one of the three crossings resolves, so the direction is a null and not a finding: what is measured here is that a great deal moved and the totals did not.

5. Where the length runs outthe same 2,048 tokens · two ways to spend them

The dial flattened at 43%, and 51% of these problems never came right at any length. But a budget is not a length: the same 2,048 tokens can be one long chain or 16 short ones, and page 7 turns on the difference.

one chain of 2,04843% solvedthe same 2,048 tokens, spent differently16 chains of 1287% each0512102415362048tokens spent · 2,048 either way
one chain of 2,04843% solved 16 chains of 1287% eachbar length is the token budget; both rows spend 2,048.
The dial has a far end, and you have reached it. This is inference-time scaling in its simplest form: the curve flattens, and 51% of these problems are not reached at any length we ran. A problem this chain cannot solve, it cannot solve by being longer. A different chain is another matter. Refusing to let this same model stop moved 30% of its answers without improving them: there is more than one answer in there, and one run only ever shows you one of them. So spend the same 2,048 tokens on 16 chains of 128 instead of one long one, sampled rather than greedy so they can differ. Page 7 asks what 16 voters at 7% each are worth.