OpenAI Locked GPT-5.6! Now what? (for trading) — Transcript

Full transcript

1. 0:00OpenAI just released GPT-5.6
2. 0:03and at least according to the trust me
3. 0:05bro benchmarks, it is even beating
4. 0:07Claude Metas 5. So, this is absolutely
5. 0:10brilliant. And not only that, more
6. 0:11competition always means lower prices,
7. 0:14which is great for us retails. However,
8. 0:16the US government has started being such
9. 0:19a buzzkill and enforced them to restrict
10. 0:22this release. That means only a limited
11. 0:24partners who have already shared their
12. 0:26documents and stuff with the government
13. 0:28are going to get access to this. In
14. 0:30other words, for us retails, we're going
15. 0:31to have to wait a lot longer to get our
16. 0:33hands on this model and start playing
17. 0:36with it. Now, if you've been following
18. 0:37my channel, then you know that what I
19. 0:39personally care is the performance of
20. 0:41these models for doing algo trading
21. 0:43research. And I always review these
22. 0:45models with different scenarios and show
23. 0:47you guys exactly how they are
24. 0:49performing. So, it was really such a
25. 0:50bummer for me not being able to do that
26. 0:53for GPT-5.6,
27. 0:54at least for the time being. But that
28. 0:56got me thinking, how much exactly do we
29. 0:58need these models? Because after the
30. 1:00release of the JCMCP, the intelligence
31. 1:03of the model itself is not as important
32. 1:06as before. Because unlike some other
33. 1:07people who are just simply farming for
34. 1:09views on YouTube and telling you that if
35. 1:12you let these models to trade by
36. 1:14themselves, you're going to make huge
37. 1:15amount of money or you're going to print
38. 1:17money, I never actually do that.
39. 1:19Instead, I use the MCP tool which can
40. 1:22perform real research. It can write the
41. 1:24strategies and then use different tools
42. 1:26such as statistical tools like rule
43. 1:28significance testing, optimization,
44. 1:30Monte Carlo simulation, and things like
45. 1:32that to ensure the code that the model
46. 1:35has just written for us has some real
47. 1:38edge. Because you never want to trade
48. 1:40some sort of black box, right? So, that
49. 1:42is why I use the MCP tool. And in my
50. 1:45personal tests, we do not need the
51. 1:48absolute latest model in order to get
52. 1:50the best results from this. So, these
53. 1:52days I am using the Opus 4.8 most of the
54. 1:55times, but sometimes when I just want
55. 1:57something to execute really fast, I just
56. 1:59set it simply to Sonnet 4.6 and it blows
57. 2:03me out of the water how good this model
58. 2:05is. So, my point is that having better
59. 2:08tools and letting it run for multiple
60. 2:10hours or days definitely beats using the
61. 2:14absolutely latest model. And if you
62. 2:16watch my latest video where I showed you
63. 2:18my 24/7 setup with Claude Code, then you
64. 2:20know that it is possible to perform
65. 2:23research for multiple days. But the
66. 2:25thing that mattered the most in this
67. 2:26video, it wasn't the model itself, it
68. 2:29was the prompt that I gave it. And
69. 2:30specifically, it was really helpful to
70. 2:33specifically tell the model that, "Hey,
71. 2:35you're going to have to run hundreds of
72. 2:37backtests for me." And that was the
73. 2:39keyword so the model won't stop early.
74. 2:41So, the point that I'm trying to make is
75. 2:43that we don't always need the absolutely
76. 2:45latest model. Sure, sometimes like in
77. 2:47this video, I showed you guys that a
78. 2:49latest model such as Fable 5 is really
79. 2:52helpful and in this case for example,
80. 2:54the Fable 5 model was able to improve
81. 2:56the speed of backtesting for us up to
82. 2:58four times, which is really huge if
83. 3:00you're trying to run any sort of
84. 3:02optimization or machine learning stuff.
85. 3:04So, yeah, it could be really helpful,
86. 3:06but once we do that, then for everyday
87. 3:08work when you want to do research for
88. 3:10trading, then we are at a point where we
89. 3:13do not need the absolutely latest model.
90. 3:15So, even Sonnet 4.6 or Claude Opus is
91. 3:18more than enough. But I want to even go
92. 3:20one step further and to claim that we
93. 3:22don't even necessarily need Anthropic
94. 3:25plans at this point. So, if you are on a
95. 3:27budget, then there are many Chinese
96. 3:29companies out there that are giving us
97. 3:31definitely good enough performance which
98. 3:33can work great with our MCP setup. So,
99. 3:36for example, GLM 5.2 was recently making
100. 3:39a lot of hype and I also tried it and it
101. 3:42works really great and it starts from
102. 3:44simply 18 bucks. However, the problem
103. 3:47that I had with them was that the speed
104. 3:49of token per second was not great in it.
105. 3:52So, that's just something to be aware
106. 3:54of. And by the way, guys, this video is
107. 3:56not sponsored by GLM or any other
108. 3:58company out there. So, this is just my
109. 4:00personal opinion about these models.
110. 4:03Now, another company is MiniMax, and
111. 4:06their coding plans begin from $20. And
112. 4:09by the way, guys, remember that all
113. 4:11these Chinese companies will give you
114. 4:13significantly higher limits per the
115. 4:15money that you are paying them comparing
116. 4:16to something like OpenAI or especially
117. 4:19Anthropic. And lastly, and definitely
118. 4:21not least, in fact, my favorite so far
119. 4:23is the Kimi Code plan. This is a company
120. 4:26that is releasing models that even
121. 4:28Americans are copying from. So, for
122. 4:30example, the Keras models, which are
123. 4:32named Composer model, is trained off of
124. 4:35this model. And if you want to get it
125. 4:37directly, it's going to begin from $19
126. 4:40per month. And their other plans are
127. 4:42also good. You see, we have the $39 per
128. 4:45month plan, for example, which is
129. 4:46something that doesn't exist with
130. 4:48Anthropic or OpenAI. So, they have the
131. 4:50$20 plan, and then they jump into $100.
132. 4:54There isn't anything in between, which
133. 4:56is, I think, something that could be
134. 4:58really helpful. Like, for example, for
135. 5:00my use case, this would be more than
136. 5:02enough. So, anyway, guys, this was a
137. 5:03short video. I just wanted to let you
138. 5:05guys about what's going on with OpenAI,
139. 5:07and also the alternative methods that
140. 5:10you can use to get the job done.
141. 5:12Because, again, we do not need to even
142. 5:14wait for these things to come out until
143. 5:16we can continue with our research. And I
144. 5:18believe at this point, people who have
145. 5:20the best tools have the real advantage.
146. 5:23So, it's not really as important as what
147. 5:25the model can write for you, it's the
148. 5:27tools that you give the model. Now, this
149. 5:29was a new type of video that I just did.
150. 5:31If you enjoyed it, please like the
151. 5:32video, and let me know what you think in
152. 5:33the comments, and subscribe to the
153. 5:35channel if you haven't already, because
154. 5:37I release algo trading tutorials all the
155. 5:39time. Thank you so much for watching.
156. 5:40I'll see you in the next one.
157. 5:45>> [music]
158. 5:51>> Mhm.