1
00:00:00,000 --> 00:00:04,480
Alright, so, I was- I was tweaking some
deployment scripts yesterday and realized my

2
00:00:04,524 --> 00:00:09,040
terminal was suddenly pulling down Claude
Code v2.1.197.

3
00:00:09,300 --> 00:00:12,480
And mate, this isn't just a minor patch.

4
00:00:12,624 --> 00:00:16,800
They have made Sonnet 5 the absolute
default model now.

5
00:00:16,933 --> 00:00:23,205
We are talking a native one-million-token
context window right in our local

6
00:00:23,272 --> 00:00:26,165
shell, brought to you by Jellypod AI.

7
00:00:26,442 --> 00:00:28,202
A million tokens natively?

8
00:00:29,086 --> 00:00:30,286
That is wild.

9
00:00:30,566 --> 00:00:34,286
I mean, you can literally feed a massive
repo into the prompt now.

10
00:00:34,446 --> 00:00:39,886
But wait, at a million tokens, the API
bill is going to hurt if you aren't careful.

11
00:00:39,966 --> 00:00:41,566
What's the damage on the pricing?

12
00:00:41,874 --> 00:00:45,234
Well, they've got this promotional pricing
running through August thirty-first,

13
00:00:45,554 --> 00:00:46,514
2026.

14
00:00:46,647 --> 00:00:51,314
It's two dollars per million input tokens,
and ten dollars per million output.

15
00:00:51,607 --> 00:00:55,714
It's- it's dirt cheap for what it is, but,
you know, only if you run it right.

16
00:00:55,901 --> 00:01:00,194
To get started, you just run npm install
-g

17
00:01:00,434 --> 00:01:07,394
@anthropic-ai/claude-code, and then you
can do /model to verify or switch

18
00:01:07,426 --> 00:01:08,834
over to sonnet-5.

19
00:01:09,014 --> 00:01:14,114
But James, there is a massive catch with
how this thing handles that giant context.

20
00:01:14,803 --> 00:01:15,698
Let me guess.

21
00:01:15,744 --> 00:01:17,018
Prompt caching.

22
00:01:17,165 --> 00:01:21,218
If you aren't hitting the cache, every
single CLI command is going to re-parse the

23
00:01:21,275 --> 00:01:22,498
entire codebase.

24
00:01:22,594 --> 00:01:25,058
That's- that is going to add up fast.

25
00:01:25,360 --> 00:01:26,320
Spot on, mate.

26
00:01:26,699 --> 00:01:27,846
It's exactly that.

27
00:01:27,983 --> 00:01:32,486
Claude Code relies heavily on prompt
caching to keep those input costs low.

28
00:01:32,666 --> 00:01:38,486
But here's the kicker: if you touch
high-level files, like- say you modify your root

29
00:01:38,546 --> 00:01:43,526
package.json or a main config file, it
invalidates the entire cache.

30
00:01:43,766 --> 00:01:48,966
It forces a complete, from-scratch reload
of that entire million-token context.

31
00:01:49,110 --> 00:01:53,286
Your next prompt will burn through the
full input cost instead of hitting the cheap

32
00:01:53,343 --> 00:01:54,246
cached rate.

33
00:01:54,553 --> 00:01:55,193
Oh, wow.

34
00:01:55,220 --> 00:01:59,433
So if you're doing a big refactor and
constantly touching the package manifest,

35
00:01:59,453 --> 00:02:00,633
you're basically burning money.

36
00:02:00,773 --> 00:02:03,353
And there's also the agentic loop issue,
right?

37
00:02:03,493 --> 00:02:08,153
Sonnet 5 is incredibly capable, but if you
let it run wild with bash commands...

38
00:02:08,676 --> 00:02:09,748
Oh, don't get me started.

39
00:02:09,881 --> 00:02:13,108
It gets itself into these recursive
bash-debugging loops.

40
00:02:13,268 --> 00:02:18,708
It'll run a test, fail, write a fix, run
the test again, fail differently,

41
00:02:18,748 --> 00:02:19,988
and just keep looping.

42
00:02:20,038 --> 00:02:24,148
Because it has that massive context
window, it doesn't run out of memory,

43
00:02:24,175 --> 00:02:27,428
it just keeps burning through tokens while
you're off making a coffee.

44
00:02:27,528 --> 00:02:30,628
You have to keep a real close eye on those
agentic runs.

45
00:02:31,510 --> 00:02:35,878
Yeah, a self-correcting loop that eats a
hundred thousand tokens a minute is a very

46
00:02:35,942 --> 00:02:37,638
expensive way to find a typo.

47
00:02:37,798 --> 00:02:42,758
But look, outside of the raw model power,
they did ship some smaller quality-of-life

48
00:02:42,811 --> 00:02:49,318
stuff in v2.1.196 and .195 that actually
solves some daily annoyances.

49
00:02:49,430 --> 00:02:52,358
Like, did they finally fix the terminal
hijacking?

50
00:02:52,662 --> 00:02:53,222
They did!

51
00:02:53,822 --> 00:02:54,702
Thank goodness.

52
00:02:54,772 --> 00:02:59,073
You can now set
CLAUDE_CODE_DISABLE_MOUSE_CLICKS=1 as an environment variable.

53
00:02:59,131 --> 00:03:02,566
Before that, it would hijack your terminal
hover and click events,

54
00:03:02,662 --> 00:03:05,062
which was bloody annoying if you just
wanted to highlight text.

55
00:03:05,062 --> 00:03:05,062
Now you can disable it entirely.

56
00:03:05,353 --> 00:03:06,313
Thank god.

57
00:03:06,409 --> 00:03:09,033
That hover hijacking was driving me
insane.

58
00:03:09,209 --> 00:03:11,593
What about file navigation?

59
00:03:11,633 --> 00:03:15,753
Because copying and pasting paths from the
terminal into VS Code is a friction point

60
00:03:15,833 --> 00:03:17,353
I hit twenty times a day.

61
00:03:17,640 --> 00:03:19,000
Yeah, they sorted that too.

62
00:03:19,768 --> 00:03:22,040
They added clickable terminal file paths.

63
00:03:22,227 --> 00:03:26,840
So now, you can just Cmd-click or
Ctrl-click right on the file path in your

64
00:03:26,893 --> 00:03:31,160
terminal, and it opens it up directly in
your IDE or file explorer.

65
00:03:31,347 --> 00:03:32,840
It's super smooth.

66
00:03:33,016 --> 00:03:37,560
Plus, if your company has specific
org-level default models set up,

67
00:03:37,600 --> 00:03:42,200
those will now actually show up properly
when you run the /model settings command.

68
00:03:42,480 --> 00:03:43,600
Okay, that's clean.

69
00:03:43,780 --> 00:03:46,480
And what about when the stream just hangs?

70
00:03:46,568 --> 00:03:50,800
Sometimes the API stalls and you're just
sitting there staring at a blinking cursor

71
00:03:50,840 --> 00:03:51,920
wondering if it's dead.

72
00:03:52,208 --> 00:03:56,128
Right, so they introduced a default
streaming watchdog.

73
00:03:56,341 --> 00:04:01,488
It's- it's enabled by default via
CLAUDE_ENABLE_STREAM_WATCHDOG=1.

74
00:04:01,621 --> 00:04:06,608
Basically, if the terminal goes five
minutes without receiving a single event from

75
00:04:06,628 --> 00:04:12,288
the API, the watchdog kicks in, times out,
and retries the connection automatically

76
00:04:12,368 --> 00:04:14,448
instead of just hanging forever.

77
00:04:14,740 --> 00:04:18,740
Five minutes is a bit long to wait, but
hey, it beats a silent death.

78
00:04:18,843 --> 00:04:24,900
Sounds like v2.1.197 is a massive step
forward if you respect the cache boundaries.

79
00:04:25,208 --> 00:04:25,848
Absolutely.

80
00:04:25,976 --> 00:04:28,968
Just watch your root configs, or your
wallet will feel it.

81
00:04:29,075 --> 00:04:30,648
Catch you on the next one, mate.

