1
00:00:00,080 --> 00:00:01,220
Here's a small mystery.

2
00:00:01,700 --> 00:00:06,240
You update Codex to version 0.159.1.

3
00:00:06,840 --> 00:00:08,600
You change nothing in your config.

4
00:00:09,220 --> 00:00:11,260
And your sessions start on a different
model.

5
00:00:12,180 --> 00:00:13,920
No banner, no prompt.

6
00:00:14,520 --> 00:00:15,060
Which model?

7
00:00:15,645 --> 00:00:16,505
Okay, I'll bite.

8
00:00:16,625 --> 00:00:18,605
Is it, um, one of the Sol ones?

9
00:00:18,917 --> 00:00:21,237
GPT 6.1 Sol.

10
00:00:21,424 --> 00:00:27,397
If config.toml has no model key, it used
to boot GPT 6 Astra,

11
00:00:27,437 --> 00:00:33,717
and as of 0.159.1 it boots GPT 6.1 Sol.

12
00:00:33,984 --> 00:00:38,597
Oh, and quick thanks to Jellypod for
helping make a daily show actually possible.

13
00:00:38,773 --> 00:00:39,677
Okay.

14
00:00:39,717 --> 00:00:40,277
Back to it.

15
00:00:40,652 --> 00:00:42,552
Wait, where did that change happen?

16
00:00:42,917 --> 00:00:49,477
The release notes list two pull requests,
49323 and 49342.

17
00:00:49,637 --> 00:00:56,517
The second one covers the Amazon Bedrock
catalogs, so GPT 6.1 Sol also

18
00:00:56,588 --> 00:01:02,357
replaces GPT 6 Sol as the default in the
Bedrock Mantle and Runtime catalogs.

19
00:01:02,497 --> 00:01:08,197
And according to a community guide to the
CLI, the model picker now lists it first,

20
00:01:08,277 --> 00:01:11,717
as the latest workhorse model for coding
and everyday work.

21
00:01:11,950 --> 00:01:15,330
Fine, but a new default is just, you know,
a new default.

22
00:01:15,530 --> 00:01:16,390
Why does anyone care?

23
00:01:16,767 --> 00:01:17,047
Money.

24
00:01:17,667 --> 00:01:24,427
OpenAI's changelog says GPT 6.1 Sol offers
near Astra performance for complex

25
00:01:24,487 --> 00:01:26,107
work at a lower cost than Astra.

26
00:01:26,807 --> 00:01:28,187
The guide puts numbers on it.

27
00:01:28,807 --> 00:01:34,347
50 credits per million input tokens, 2.5
cached, 250 output.

28
00:01:35,107 --> 00:01:39,767
Astra is 250, 25, and 1,250.

29
00:01:40,092 --> 00:01:40,572
Hold on.

30
00:01:40,752 --> 00:01:43,232
So that's a, a fifth of the price.

31
00:01:43,702 --> 00:01:44,862
Eighty percent off.

32
00:01:45,182 --> 00:01:45,482
Yeah.

33
00:01:47,338 --> 00:01:48,758
Okay, that's not nothing.

34
00:01:48,917 --> 00:01:51,317
But the sneakier part is the service tier.

35
00:01:51,457 --> 00:01:57,477
The same guide says GPT 6 Sol and Luna
ship with Fast as the default tier.

36
00:01:57,717 --> 00:02:03,157
On ChatGPT sign in, Fast bills at 2.5
times Standard credits.

37
00:02:03,344 --> 00:02:08,037
So a session where you never set a service
tier could quietly start in Fast.

38
00:02:08,402 --> 00:02:11,342
So people were paying a tax they didn't
know existed.

39
00:02:11,625 --> 00:02:12,745
Possibly, yeah.

40
00:02:13,065 --> 00:02:17,065
GPT 6.1 Sol has no default service tier.

41
00:02:17,172 --> 00:02:19,385
Sessions start at Standard speed.

42
00:02:19,525 --> 00:02:22,665
You can check with the slash fast status
command.

43
00:02:23,207 --> 00:02:23,507
Huh.

44
00:02:24,127 --> 00:02:27,507
That's, that's honestly a bigger deal than
the price table.

45
00:02:27,967 --> 00:02:28,867
The price you can see.

46
00:02:28,907 --> 00:02:30,067
The tier you can't.

47
00:02:30,417 --> 00:02:31,217
Right.

48
00:02:31,237 --> 00:02:32,497
And there's a third change.

49
00:02:32,607 --> 00:02:37,937
Default reasoning effort is low, where GPT
6 Sol was medium.

50
00:02:38,298 --> 00:02:40,118
Oh, I felt that one.

51
00:02:40,698 --> 00:02:45,618
I ran it on a multi file refactor, and
turns came back in seconds.

52
00:02:46,218 --> 00:02:49,418
The reasoning was, um, really sparse.

53
00:02:49,838 --> 00:02:51,698
Almost suspiciously quick.

54
00:02:52,000 --> 00:02:53,600
Was the result any good?

55
00:02:55,445 --> 00:02:59,465
For mechanical stuff, renames, moving
things around, yes.

56
00:03:00,065 --> 00:03:03,425
But if the job is deciding how the pieces
should fit together,

57
00:03:03,965 --> 00:03:04,885
I'd raise the effort.

58
00:03:05,565 --> 00:03:09,745
Medium, high, and the SDK notes even
mention an ultra value.

59
00:03:10,125 --> 00:03:15,045
So cheaper, slower to burn credits, and
thinks less out of the box.

60
00:03:15,133 --> 00:03:16,125
That's the trade.

61
00:03:16,425 --> 00:03:19,265
A fair trade, as long as you know you're
making it.

62
00:03:19,625 --> 00:03:20,425
Now the catch.

63
00:03:20,569 --> 00:03:24,505
Does upgrading the CLI move you to 6.1
Sol?

64
00:03:24,982 --> 00:03:25,942
I'd have guessed yes.

65
00:03:26,662 --> 00:03:27,782
New version, new model.

66
00:03:28,125 --> 00:03:28,805
Nope.

67
00:03:28,875 --> 00:03:33,005
The default only applies when config.toml
names no model.

68
00:03:33,218 --> 00:03:39,085
If you pinned model to GPT 6 Sol, or GPT 6
Astra, you keep that.

69
00:03:39,232 --> 00:03:41,405
Saved choices aren't migrated.

70
00:03:41,737 --> 00:03:42,777
And the older stuff?

71
00:03:43,077 --> 00:03:44,697
Like GPT 5.6?

72
00:03:45,042 --> 00:03:50,562
Per the guide, the migration prompts for
the older picks, GPT 5.6 Sol,

73
00:03:50,922 --> 00:03:57,922
GPT 5.6 Terra, GPT 5.5, still point to GPT
6 Sol.

74
00:03:58,102 --> 00:03:59,962
Not 6.1.

75
00:04:00,082 --> 00:04:05,042
So somebody can click through a migration
and land one generation behind without

76
00:04:05,104 --> 00:04:05,682
noticing.

77
00:04:06,108 --> 00:04:06,268
Ugh.

78
00:04:06,868 --> 00:04:12,548
So to actually get it, you set the model
key yourself, or pass the m flag on the

79
00:04:12,588 --> 00:04:13,048
command line.

80
00:04:13,417 --> 00:04:13,817
Yep.

81
00:04:13,929 --> 00:04:14,857
Next gotcha.

82
00:04:15,001 --> 00:04:20,697
Rollout covers Plus, Pro, Business,
Enterprise, and Edu.

83
00:04:20,837 --> 00:04:25,497
But Enterprise and Edu keep it off until
an administrator enables it.

84
00:04:25,689 --> 00:04:28,617
Free and Go aren't included at launch.

85
00:04:28,985 --> 00:04:32,785
So the whole team upgrades, the admin
hasn't flipped the switch,

86
00:04:32,985 --> 00:04:35,685
and half of them are wondering why nothing
changed.

87
00:04:36,042 --> 00:04:38,362
Or why the model they expected isn't
there.

88
00:04:38,575 --> 00:04:40,242
So verify.

89
00:04:40,329 --> 00:04:47,082
Slash status, which as of 0.159.0 renders
without borders and

90
00:04:47,135 --> 00:04:53,162
wraps long values, or open the slash model
picker and see whether 6.1 Sol is even

91
00:04:53,231 --> 00:04:53,722
listed.

92
00:04:54,050 --> 00:04:56,110
And if it's not, pin one you do have.

93
00:04:56,500 --> 00:04:57,260
Right.

94
00:04:57,330 --> 00:04:58,820
Now, the patch train.

95
00:04:59,140 --> 00:05:05,460
0.159.2 landed the same day as 0.159.1,

96
00:05:05,580 --> 00:05:07,820
about three and a half hours later.

97
00:05:07,940 --> 00:05:14,820
It's one backport, pull request 49385, and
it stops Windows from flashing

98
00:05:14,900 --> 00:05:19,220
a console window every time Codex starts a
background process.

99
00:05:19,790 --> 00:05:20,690
Oh, the flashing.

100
00:05:21,170 --> 00:05:23,830
Shell commands, hooks, Git calls, all of
it?

101
00:05:24,292 --> 00:05:29,972
The guide says that covers shell tool
commands, hooks, Git and npm calls,

102
00:05:30,063 --> 00:05:33,572
doctor checks, sandbox setup refreshes.

103
00:05:33,670 --> 00:05:36,692
Commands that need your terminal still use
it.

104
00:05:36,745 --> 00:05:37,972
No defaults change.

105
00:05:38,420 --> 00:05:42,980
Small fix, but if you've watched windows
pop up forty times during a task,

106
00:05:43,340 --> 00:05:44,160
you remember it.

107
00:05:44,500 --> 00:05:49,540
Then 0.159.3, on September 30th.

108
00:05:49,700 --> 00:05:53,140
Pull request 49744.

109
00:05:53,280 --> 00:05:59,380
The notes say eligible local sessions
signed in with ChatGPT can now show optional

110
00:05:59,444 --> 00:06:02,020
reminders to complete account security
setup.

111
00:06:02,462 --> 00:06:02,962
Optional.

112
00:06:03,342 --> 00:06:05,062
So a nudge, not a wall.

113
00:06:05,333 --> 00:06:06,413
A nudge.

114
00:06:06,503 --> 00:06:10,693
And the clock to watch is October 14th,
2026.

115
00:06:10,821 --> 00:06:16,613
That's when GPT 5.5 retires from Codex for
ChatGPT sign in.

116
00:06:17,902 --> 00:06:23,622
So if your config still says 5.5, you've
got until then to choose on purpose.

117
00:06:24,042 --> 00:06:24,562
Yeah.

118
00:06:24,595 --> 00:06:25,682
Check the config.

119
00:06:25,722 --> 00:06:26,602
That's it from me.