OpenAI / o200k_base · models.dev

GPT-5.6 Sol

Opened from a screened package ae394ceb489c.

Encoding fertility 1.21×

Harnesses

Same model, different drivers. ChatGPT is not OpenCode. $ / MU is only set when every official task passed. $ / pass still shows on a partial run; it does not rank. Time is output tokens at 80 tok/s, not wall clock.

Stack	Passed	Attempts	Time	$ / MU	$ / pass	$ / hr	Tokens / pass	Burn	$ / OpenAI
ChatGPTxhigh	3/3	4	7m 39s	$0.000584	$0.8351	$13.95	73,382	1.18×	$0.8351
ChatGPTultra	3/3	10	27m 47s	$0.002971	$4.25	$14.91	255,756	4.12×	$4.25
ChatGPTmax	3/3	4	9m 02s	$0.000719	$1.03	$14.63	86,928	1.40×	$1.03
ChatGPThigh	3/3	5	5m 39s	$0.000483	$0.6913	$14.10	62,123	1.00×	$0.6913

Official jobs

Pass or fail on each sealed job. Attempts include retries. Failed tries stay in the bill.

Harness	Durable work queue 4/4	JSON Patch (RFC 6902 subset) 4/4	Token bucket and sliding window 4/4
ChatGPT · xhigh	pass · 2	pass · 1	pass · 1
ChatGPT · ultra	pass · 8	pass · 1	pass · 1
ChatGPT · max	pass · 2	pass · 1	pass · 1
ChatGPT · high	pass · 3	pass · 1	pass · 1

Tokens by job

Billed tokens on each official job, including retries. Cache reads drop out of this count.

ChatGPT · xhigh
122,698Durable work queue
39,036JSON Patch (RFC 6902 subset)
58,413Token bucket and sliding window
ChatGPT · ultra
532,594Durable work queue
74,798JSON Patch (RFC 6902 subset)
159,875Token bucket and sliding window
ChatGPT · max
174,422Durable work queue
52,033JSON Patch (RFC 6902 subset)
34,329Token bucket and sliding window
ChatGPT · high
133,180Durable work queue
24,190JSON Patch (RFC 6902 subset)
28,999Token bucket and sliding window

Where the bill goes

Failed attempts stay in the bill. Cache hits are a discount, shown separately.

ChatGPT · xhigh220,147
Cache hit 972,800 (discount, not extra work)
ChatGPT · ultra767,267
Cache hit 11,171,584 (discount, not extra work)
ChatGPT · max260,784
Cache hit 1,393,920 (discount, not extra work)
ChatGPT · high186,369
Cache hit 926,720 (discount, not extra work)

Uncached in
Output
Thinking
Cache write

Retry tax

Attempts are harness calls, including failed hidden-verifier retries. Extra tries stay in the bill.

ChatGPTxhigh4 attempts · 1 extra / 3 jobs
ChatGPTultra10 attempts · 7 extra / 3 jobs
ChatGPTmax4 attempts · 1 extra / 3 jobs
ChatGPThigh5 attempts · 2 extra / 3 jobs

Cache hits

Hits divided by uncached input plus hits. No cache means the harness did not report a discount.

ChatGPTxhigh84%
ChatGPTultra95%
ChatGPTmax87%
ChatGPThigh85%

Time vs $ / MU

Decode time at 80 tok/s against what that finish cost. A long token bill is not the same as an expensive one.

Encoding by slice

Native tokens per MU on the same frozen basket. Tool schemas are where fertility often doubles.

English prose30% of the basket0.93×
Code25% of the basket1.00×
Structured15% of the basket0.85×
Tool schemas10% of the basket1.03×
CJK10% of the basket3.61×
Instructions10% of the basket0.90×

Slice	Characters	Native tokens	Fertility	True in	Source
English prose	4,474	1,039	0.93×	$4.64	official
Code	2,321	581	1.00×	$5.01	official
Structured	3,067	649	0.85×	$4.23	official
Tool schemas	2,378	613	1.03×	$5.16	official
CJK	831	751	3.61×	$18.07	official
Instructions	1,164	261	0.90×	$4.48	official

Endpoints

Provider	SKU	List in	List out	Cache hit	Cache write
OpenAI	`gpt-5.6-sol`	$5.00	$30.00	$0.50	$6.25

All stacks