シャドーイング練習: OpenAI fights back - 動画で英語スピーキングを学ぶ

they're back okay let me catch you guys up quick last week anthropic dropped a new model opus 5 .5

and it was unbelievably good it was so unbelievably good

that openai rushed out two model drops gpt6 soul

and gpt6 luna you might have noticed i didn't do a video on those models there's a reason they weren't

that good i was not particularly impressed with either of them

and didn't really have much to say

but there was one other model i happened to get early access to

that is now available for all that model is called gpt 6 .1 sol

and this model made it very very hard to film

that sonnet 5 .5 video because i knew something was coming something surprisingly cheap surprisingly capable

and most surprisingly better than astra

so yeah i got a lot to say about this one i'm filming this at two in the morning, right after filming my Sonnet 5 .5 videos.

So pardon me for stumbling over a few words here and there.

I'm doing my best to get this out as reasonably quickly as possible because I want to have some coverage.

And I'll be real, it is also quite fun to cover these things before they are out.

So you're getting my true honest take and not the distilled version of what everyone else is saying.

I'm sure this model is going to cause some pretty crazy waves.

So it'll be nice to have my take out initially separately first.

As always, I feel obligated to remind you i do have early access

but i'm not being paid in any way shape

or form openai has no influence over what

and how i say things just

when they politely asked me to wait until the model is out to talk about it

which makes a lot of sense

but i have to wait for one other thing first today's

sponsor in order to build good software with agents they need to get feedback

and let's be real they're getting a lot of

that feedback from our ci that's why we've all been seeing our ci bills skyrocket

and also why we've been getting more

and more frustrated with github actions today's sponsor is depot

and they're here to solve all of this and more Not only can they make your CI up to 10 times faster, as well as your Docker builds up to 40 times faster, especially when they're downloading cache,

they're also cheaper, and they give better feedback for your agents.

All this is possible due to Depot Metal.

They're running their own bare metal with AMD EPYC processors

that are way faster than what you get from traditional CI providers like, of course, GitHub Actions.

If you want it to be a drop -in replacement, it absolutely can be, but Depot's APIs are so much better

that you should probably use those instead they enable parallelization

and most importantly resilience

when github inevitably goes down randomly for no good reason we've had our releases get blocked

because we weren't using depot and i'm so thankful

that i've been moving more and more stuff over for example

when ben moved picthing over to bun we immediately had some

ci failures normally this would be obscure piles of text

that our agents parse through for us but

when we use depot it becomes way easier to see they'll even analyze the failures

and give suggestions to make it much simpler to get this feedback back to our agents this is especially useful

when you tell your agents that you can use depot

because they'll no longer have to push changes and wait for

that to trigger a build they can just run the cli to trigger the exact same ci

that you'd be triggering through github instead no longer do you

have to file prs with broken code just to get feedback

to your agents they could just run a tool instead your

agents will also get way better breakdowns of what is taking

so long in your actual ci runs so

that you can figure out how to improve them and make them faster and more reliable.

You and your agents deserve faster Docker, faster build times, faster CI, better results, and ideally a cheaper price.

Get all of that and more at soydiv .link.

Let's talk about this model a bit

because it is not quite what I expected and it's probably not what you guys expected either.

Especially when you consider the GPT -6 sole just came out like a week ago.

It'll be around a one week gap from 6 .0 sole to 6 .1 sole.

I also want to disclose the numbers i'm currently showing on my screen are unlikely to be exactly accurate

because i am running terminal bench for myself in the first

two times i ran it i screwed things up the third

one seems to be doing much better i didn't run it on medium initially

so there's a miss it there but low high x high

and max although the max run is incomplete

so i'm currently backfilling scores from x high for the ones

that max either got wrong or in a previous run

or didn't do yet because it takes like eight plus hours on some of these tasks.

This bench is nuts.

I'm more skeptical of benchmarks than ever now

that I've been running a lot more of them myself in order to get the coverage I want to give here.

For what it is worth, Terminal Bench 4 is state -of -the -art score here, as is Deep SWE, although this one's weirder

because it goes down on X high and max and stays even on low and high, but those even low and high scores are scoring around what Astra did on high.

The difference being it's doing it for comically cheaper switch over

to the log scale you'll see what i mean this model

on low costs 21 cents versus aster on low costing a dollar 46

and opus 5 .5 on max getting the same score for 14 .65

while i will gladly admit

that deep sweet is far from a perfect measure of how

good a model is at day -to -day code work the fact

that 6 .1 soul is scoring the same as opus and is also 73x cheaper.

is at least worth noticing.

Here's where I'm going to say some of the things that I probably shouldn't.

Considering that GPT -6 Sol came out last week on Tuesday, and this model's coming out this week on Tuesday,

I think it's reasonable to infer that 6 .1 Sol was not meant to be 6 .1 Sol.

There are things I'm not supposed to say, and I'm definitely walking a thin line here by sharing it, so I hope this proves I'm not paid off by OpenAI,

because I'm about to give you guys info that I.. yeah, just let me get through this.

First and foremost, 6 .1 is significantly smarter of the GPT -6 Sol, thereby indicating this isn't just a .1 bump, there is something fundamentally different here.

Next point is that it has meaningfully slower tokens per second.

That tends to indicate the model is bigger, hard to know for sure.

Seems like this model might be different.

Most importantly, we have Tebow's tweet.

What am I referring to there?

Well, right before I started filming, Tebow dropped quite a wall of text.

The thing I want to emphasize here is first off that the pro $200 subscription is back.

But more importantly, more importantly is this sentence, they are changing how they calculate the usage in the sub.

In effect, if you do the math, it will net out at half the dollar in API spend compared to the old pro $200 plan.

100

Why in the world would they do this, especially right now where there's allegedly an internal code red

101

because opus 5 .5 is so unbelievably good

102

and has made the 200 cloud code sub such an unbelievable

103

value the only reason in the world tebow would post this

104

right now is uh i don't know maybe a new model is coming where their margins aren't as good

105

so the ability to subsidize has gone down

106

because remember you can get eight to nine thousand dollars of usage in a month on the 200

107

cloud code plan

108

and you can get over 12 grand on the 200 codex

109

plan i did actually run a lot of numbers before this

110

and the amount you could get on Astra did go down slightly closer to like eight grand

111

or so hard to know for sure because they differ for everyone everywhere

112

and it's hard to log all of this stuff

113

but for my math roughly nine grand a month the usage

114

and that's where the price for this model comes in this model is two dollars per million input tokens

115

and ten dollars per million out this makes it way cheaper

116

than 5 .6 soul was at launch half the price of 5 .6 Sol after discounts

117

and the same price as GPT -6 Sol, one -fifth the price of Astra.

118

However, this is not the whole story because cash reads matter.

119

And the cash read price for this model is going to be 10 cents per mil in.

120

That's a big deal.

121

OpenAI has not changed cash read price.

122

As far as I know ever before, it's always been exactly 10 % of the normal read price.

123

That makes it a 90 % discount.

124

And now it's a 95 % discount.

125

That means they cut the cash read cost in half massively reducing the cost for real world agentic use

126

which to be clear is what we're using these for most of the time

127

so this makes the model absurdly cheap for doing real world code work

128

that also means

129

that they are almost certainly cutting into their margins historically these

130

margins are rumored to be as high as 95 percent like

131

for every 10 you spend they only have to spend 50 cents

132

and as crazy as that sounds it makes a lot of sense especially

133

when you consider how expensive it is to make and train these models.

134

But that also gives them wiggle room to change things around a bit, which appears to be what's happening here.

135

That also means that if they were to keep subsidizing the same level that they were on the subscriptions, that your electricity cost for your sub would be more than you're paying.

136

So I get why they have to change this.

137

They've kind of just left the details out there for us to reverse engineer.

138

So take this as you will 6 .1 coming

139

so fast seems to indicate it is not just a new snapshot of gpt6

140

so let's talk more about this model as i was showing

141

earlier seems really good at terminal bench every time we refresh the numbers change

142

because new runs come in and it looks like max failed some things

143

that x high passed which is why i just dropped a bit

144

but again pretty much all of these even high

145

and x high are scoring higher than anything else ever has

146

and this is for me running this benchmark on random VMs on my network.

147

So not the best suite to test against.

148

I also had to drop three particular tasks from it because they expected an H100 to work against, which I make decent money.

149

I don't make H100 money, okay?

150

But none of this is real world code work.

151

So let's talk a bit about that.

152

Obviously, we'll have all the fun things like fish slap near the end.

153

So stay tuned for that.

154

But I just want to fixate a bit on the costs here.

155

Because the most expensive run with 6 .1 sol for me was about $1 .38 per task.

156

And the cheapest run with Opus 5 .5 was $5 .12.

157

That's a 4 to 5x gap from the cheapest Opus to the most expensive Sol.

158

So at this point, I would imagine you are hoping

159

and praying this model is good and that it can actually replace Opus 5 .5 for day -to -day work.

160

And I promise we'll get some good answers to that in a bit.

161

But first, we need to talk a bit about model behaviors here, because this model is a part of the GPT -6 family,

162

which means it has behaviors that are worth talking about.

163

I know I cite this diagram a lot, but there's a reason for it.

164

The thing that made me so frustrated with GPT -6 Astra wasn't

165

that it was less intelligent than the best models from Anthropic, because it was more intelligent than the best models from Anthropic, and I would argue in many ways still is.

166

But there is a problem.

167

It is also dumb.

168

It is smart and dumb at the same time.

169

GWT -6 Astra would just randomly spike into the dumbest bullshit I've seen a model do this year, even worse than some of the small open -weight models I play with.

170

It's still so deeply frustrating that Astra does this, because on the other end, when it does well, it's unbelievable.

171

But these spikes got to the point where I effectively churned.

172

I was only using my codec subs for computer use, and I ended up just leaning on to Fable 5 .1

173

and obviously now Opus 5 .5 for almost all of my day -to -day work.

174

So have they addressed the spikiness?

175

Has GPT 6 .1 Sol fixed the problems that I was so frustrated about with Astra?

176

I would say mostly.

177

Not entirely, but for the most part, yeah, this is a much better model.

178

Its peaks are not as high.

179

This is not the incredible revolutionary 3D capabilities that we saw with Astra.

180

In fact, I would put it slightly below 5 .5 Opus in most of those types of things.

181

It is not as good at computer use as Astra, although it is close enough to the point where I have

182

been happy using it for all of my day -to -day work.

183

I actually had 6 .1 Sol go through all of my emails

184

and find invoices that I had forgotten to pay or was behind on, mostly like investing stuff, and set up new tabs in Chrome for every investment I needed to wire,

185

fill out all the details for me, and just leave me to hit send.

186

It didn't get a single thing wrong and called out additional stuff

187

that I absolutely would have missed if I was doing this work myself

188

so I'm literally trusting this model to wire money for me it's trustworthy enough for

189

that and honestly I don't know if I would have trusted Astra with

190

that due to the spikiness six point one soul much much

191

less spiky from what I've heard from the other testers they

192

seem to agree with this analysis I know for a fact that Julius

193

and Ben who have also been testing have had a much

194

better experience with this than Astra in terms of the spikiness

195

Julius called the model incredible Ben called it incredibly boring I

196

think that's the best place you can be for a model drop like this

197

but as i had mentioned before its peaks are not as impressive

198

while it does quality work the majority of the time there are some tasks

199

that are just at the edge of its capability

200

that it will start to do weirder stuff on for the most part it's fine

201

but i i'm still reaching for opus a decent bit we'll

202

talk more about the comparison later i do default to this

203

model for a bunch of stuff though first off as i

204

mentioned before computer use i can't wait for ultra fast to be like an actual thing you can use with OpenAI models, because when it is, this model is going to be crazy on it,

205

because it can already figure out how to navigate computer use totally fine.

206

If it can suddenly do it six times faster, it's going to be unbelievably fun.

207

Still not quite as good as Astra, but more than good enough that for the price difference, I wouldn't even think twice about it.

208

But as I mentioned before, there are certain things I would still occasionally use Astra for that I am more than happy to use Sol for.

209

One of those things is deep code reviews.

210

I have still found OpenAI models

211

and the like Rottweiler nature where they'll dig into a problem

212

and shake it and tear it to pieces until they find every single thing wrong with it.

213

I find 6 .1 Sol to be incredibly capable in this particular way.

214

So as you can probably guess, I had 6 .1 Sol do some deep audits on Orchestrator V2 and other parts of my real world code bases.

215

In my Orchestrator V2 audit, it performed nearly identically to Astra.

216

I do believe it was slightly higher a score.

217

Okay, not in this analysis, but in my other analysis, it did actually score very, very slightly higher, but it did it at about half the price, $297 versus $584.

218

Sonnet was still cheaper, and Opus was slightly cheaper as well.

219

The difference being, neither of these models were anywhere near as thorough with their analysis.

220

6 .1 Sol dug deep to find things, which is why it was able to get a score comparable to Astra, although it did admittedly burn way more tokens.

221

Another task I've had a lot of fun testing with is asking the model to find opportunities to improve a code base.

222

In this case, to improve T3 code.

223

This is the one where Grok 4 .7 scored strangely well.

224

Of course, Astro scored way better at an 83 .8 versus the 80 .7 from Grok 4 .7.

225

But GBD61 Sol hit it out of the park with an 87 .4.

226

I didn't save all the prices for these runs.

227

It's been a bit, okay?

228

But for Opus 5 .5, it cost five bucks

229

and for sonnet 5 .5 it cost almost nine dollars with gbd61 solid was two dollars

230

and 15 cents that's the difference this model's token price is cheaper than sonnet

231

but its token utilization is still maintaining openai's usual efficiency which results in just crazy price to performance.

232

This whole thread was particularly fun

233

because I had Opus 5 .5 review this model with a different name obviously so I didn't know what it was.

234

I went and edited the history after.

235

And it concluded very quickly this was a frontier tier model.

236

Its reviews and bug repros matched the fixes that later merged.

237

Its first draft code had real bugs which review bots caught.

238

Four reviewers are still running.

239

The local T3 code reviews back frontier tier again in a

240

blind 10 model bench on the same prompt 6 .1 souls

241

placed first of the a7 .4 it found the fish slop runs

242

and compared those two it did say 6 .1 souls quality

243

output was slightly below astra's as well as the two open ai models with opus 55

244

and sona 55 which we will absolutely show you in a bit

245

but i do want to call out the price here

246

because it only cost seven dollars to run versus 15 for sona 55

247

and 50 for opus opus's honest tier call was

248

that this model is incredible for scoped work top of the frontier refine what's wrong

249

and tell me the truth i would choose it over astra

250

and about level with opus for a long unattended building this was below frontier follows its process rules even

251

when they stop all progress

252

and it does not ask for help this i absolutely noticed i had mentioned before well a few times now

253

that my ts rust port

254

that i'm making with opus 5 .5 is going way better than

255

when i was working on that same port using Astra and Sol in the past.

256

I had that port running for a while with this model and it made no progress.

257

It burned a shitload of tokens, but it didn't actually improve the compiler at all.

258

Opus was able to from scratch restart it

259

and get it working in a day after I had spent months

260

and hundreds of thousands of dollars in tokens with this model as well as with Astra and 56 Sol.

261

Opus did it in like a grand in like a night with just two subscriptions with the cloud plan

262

so for unattended long like heavy rewrite type stuff anthropic is just comically far ahead right now

263

and it also didn't have great judgment

264

when i was using it for managing my fleet

265

and for those wondering my fleet is all the computers i use for running all my agents

266

and code because one computer is far from enough i don't run any of them on this macbook now

267

so when i use this model to manage the fleet it made a couple dumb mistakes here

268

and there to be fair so is opus astra is the only one

269

that hasn't really made too many of those dumb mistakes

270

but like I'm gonna be

271

so real I am entirely done using Astra for this model

272

after I had Opus do all of this review I asked

273

it how much does it think this model should cost it

274

guessed $5 per mil in 50 cents cashed in 30 per

275

mil out putting it at Opus's prices roughly it said that

276

because it's performing like Opus its speed should add a premium

277

because it is quite fast and it's not a pro model

278

which is where it expects those higher like hundred dollar out tiered pricing things to come.

279

ProModel isn't really a thing anymore.

280

We just use Fable and Aster, but you get the idea.

281

This is the funniest part of the whole thread, though.

282

If OpenAI wants people to adopt it, I would expect $3 per mil in, $0 .30 for cashed, and $20 instead.

283

That would still be a fair price for what it does.

284

To which I responded, if I told you it was $2 in, $10 out, and $0 .10 per mil cash read, what would you think?

285

I'd call that very aggressive pricing.

286

For how you use it, it costs is about a quarter of what I guessed.

287

The cash price does most of the work.

288

Agent workloads are 96 % cash reads.

289

So 10 cents for cash reads matters more than the $2 and $10 headline prices.

290

When I looked at all of my sessions, its price guess would have been $5 ,700.

291

But after looking at these new prices, it redid the math and it would have been $1 ,550.

292

That is a massive decrease.

293

And for all my PR review type tasks, it was expecting those to be up to $10 and it's actually only up to $3.

294

And that's where like heavy PRs with 10s of 1000s of lines of code, according to Opus.

295

So don't blame me blame Opus for saying this.

296

First off Opus says it becomes the default model for scoped work.

297

Second off, it says that bloated system prompts barely matter anymore because of the cash pricing.

298

It's just noise.

299

Third, it says long loops are still a bad idea, but not because of money.

300

It's because according to it, the TSRL support wasted four days and made no progress at all.

301

And Opus even said they'd be suspicious of it lasting.

302

They expect this price to go up in the future.

303

I cannot fathom OpenAI ever increasing the price for a model, but Opus thinking they will is hilarious and shows just how good a value the model is.

304

I love this call out here.

305

GBD 6 .1 Sol did three rounds of work for about half the cost of Sonnet's single round.

306

A lot of this comes down to how context was managed, both because 6 .1 Sol is much more efficient so it's not doing as many calls that bloat the context.

307

It's not outputting as many tokens that are like building up over time.

308

So the average number of tokens being read per request was only around 110 ,000 tokens versus 360 ,000 for Sonnet 5.

309

The result is that Solout used under half as many input tokens as Sonnet making this model significantly more efficient.

310

Speaking of efficiency I want to talk about these deep SWE scores a tiny bit more

311

because this is a weird bench for me to have forked

312

and include in these things i actually did it for a different reason not to compare against 6 .1 sol

313

but to compare against a new release from open router

314

jev router open router added jev router to try

315

and optimize costs with your requests

316

and i thought it was an incredibly stupid idea once i

317

started running it against benchmarks i confirmed it's an incredibly stupid idea it turns out a model

318

that cannot reason

319

that is given a prompt into no context cannot make a good decision around how hard the problem is

320

and jev router ended up being deep seek v 4 .1

321

flash router for the vast majority of its runs around 60

322

of all the requests went straight to deep seek 4 .1 flash

323

so didn't like it that much it also routes to other smarter models

324

which should give it more of an advantage

325

but it ended up being more expensive than gpt6 astro was on low

326

while also taking four to five times longer

327

because six astro low took 4 .6 minutes

328

and jev router took 20 jev router's average task took 104

329

steps whereas gpt6 astros took 19 you get the idea it wasn't very good

330

but the whole point of jev router is

331

that it would be as cheap as possible to get a certain score

332

that was the promise on the tin whether

333

or not you believe them is up to you not me

334

i think it's bullshit regardless jev router was routing to deep

335

seek 4 .1 flash for the majority of its requests despite

336

jev router routing to the cheapest possible small open weight models

337

from whatever provider will give it away for free 6 .1

338

soul on low got the same score for an eighth the price.

339

OpenAI is here to destroy any wins anyone else has in terms of efficiency.

340

Completing this bench in 4 .8 minutes for 21 cents with

341

the second highest score I've ever seen on it is a massive achievement, tying Opus 5 .5, which took 50 minutes per task on max,

342

a tenth the time and a 70th the price for the same score.

343

If your work fits within the things 6 .1 Sonnet does well, you should probably use it for everything.

344

But if your work doesn't fit in it particularly well, you should probably keep using Opus, and maybe give Opus the ability to call 6 .1 Sol when it should for various tasks.

345

I'm almost certainly going to be setting things up so

346

that Opus 5 .5 can call 6 .1 Sol to do investigation work to try and root cause bugs,

347

to do analysis of codebases to figure out what things need to be touched and why, to help me triage real world work, to help me review the work that Opus does and more.

348

I'm kind of spoiling the ending here, aren't I?

349

I'm going to keep using Opus 5 .5 for now.

350

Before I explain why, let me do the thing that I'm most excited for.

351

Fish Slop!

352

The first thing you might have noticed is the inclusion of Slop in Fish Slop.

353

This model did the horrible thing I hate, where it surrounded the game in a bunch of absolutely garbage UI

354

and coming to this right after the 5 .5 sonnet demo hurts me deeply

355

because sonnet 5 .5 did not make graphics anywhere near this good looking

356

but at least it made a ui that was nowhere near this awful

357

and man do i wish the bad ui is where the

358

issue stopped i will turn on the sound oh god it's blaring

359

it is stunning looking the fish are some of the best

360

the model for the sub is way better the propellers work way better i'm gonna mute the sound

361

because that is looking pretty bad i haven't even heard it honestly

362

but damn like looks beautiful but

363

if you actually are playing it one of the first things you'll notice is

364

that the movement feels significantly worse than it does in either the Opus

365

or the Sonnet versions that I have demoed in the past.

366

Yeah, it moves jank.

367

It also has a significantly worse frame rate than the versions from the other models.

368

It does have higher graphic fidelity, so that makes sense.

369

The models here with the plants are significantly better than they were with the Sonnet version.

370

The derives of the fidelity of the extras in the tank is absolutely hilarious.

371

Like, yeah.

372

But goddamn, I'm so tired of the unnecessary text everywhere.

373

This model doesn't worse than almost any I've ever seen before.

374

We got take a breather, paused.

375

Your little world can wait.

376

Back to the reef.

377

Start start a new tank, slop01, feeder submarine, a little underwater chaos,

378

the big little goal, your little ecosystem, little fish become big earners, four meals, and they're all grown up.

379

Make the family a little bigger, a little golden overachiever.

380

There's so many of these.

381

There's like 20 plus of them.

382

And I promise you guys, as soon as I saw this, I took a screenshot, i sent it to open ai

383

and i crashed out in the slack

384

because i cannot fathom how they haven't fixed this problem this

385

model is unacceptably garbage at ui it has regressed again and

386

if you're looking for a model that can make front ends that don't suck

387

go spend your money somewhere else

388

because it should not be spent here this model sucks at

389

front end it sucks at design it has no taste

390

and you're gonna have to bring your taste yourself still

391

but it is admittedly really good at blender

392

if you give it like a screenshot of a thing you want it to model in 3d

393

and say hey you have blender over the cli go make

394

this it will it'll do a pretty damn good job

395

but i would never have it make the actual mechanics for my games

396

because it feels awful to play it also has like nowhere

397

near as much gameplay loop in fact the first time i

398

tried demoing this before filming it just randomly game overed as

399

i was like getting started in the first 30 seconds

400

and never like said why actually i think i technically beat it

401

i also want to like take this version and hand it to opus

402

or sonnet and say hey can you make this play better

403

because the graphics are good but the game sucks but

404

when you combine how cheap it was to make this

405

because like this was five dollars i think to generate that's pretty insane

406

and if you combine that with like ultra fast if that ever happens

407

suddenly you're going to be able to make a game in

408

a few minutes on demand we're actually now getting to

409

that threshold where game development is about to flip upside down

410

because of how models are finally understanding three -dimensional space

411

and the tooling necessary to do these types of things it's

412

happening as per usual i was not allowed to put the

413

code i wrote with this model inside of t3 code

414

or other open source projects during the testing window

415

so i had to use it exclusively on my internal projects like lakebed as well as for auditing other work

416

which means i mostly uses for auditing other work and I was very impressed.

417

This is a real PR I was working on to fix

418

a bug where my new little work tree like setup window

419

that would appear in a new thread in T3 code would disappear if you left and came back.

420

I had cloud code work on this but this problem went pretty deep

421

so I wanted to make sure that whatever solution I came up with was very very very well vetted.

422

While I personally still do not trust this model to write

423

the code I'm trying to land I absolutely trust it to review things.

424

Ignore the GPT -6 soul there, just a placeholder.

425

So when I had 6 .1 soul look through this, it found real problems that were entirely missed by Fable and by Opus.

426

First, it called out that follow -up messages can stay blocked after the agent starts, which is very annoying if you want to queue a message.

427

And also that recovered setup progress was disappearing too early.

428

It figured all of these things out with a combination of reading the code and analyzing it as well as computer use.

429

And it's able to prevent me from merging a real regression in t3 code.

430

So I literally just copy pasted those things to Claude and then told it to take another look.

431

Said better, but I'll still fix two small gaps.

432

Remember, I can't use this model of the code for this project at the time.

433

So I copy pasted that again over to Claude and it eventually got it good enough.

434

And then I finally merged.

435

But that is what I like this model for.

436

And I cannot wait to push its limits for actually coding.

437

Although I will say from the code that I did have the misfortune of reading, it is harder to justify merging this code than it is

438

for code from opus normally i would make you guys wait for the opus versus soul video

439

or the sonnet versus soul video

440

but i'll just spoil the details now i like using them in tandem

441

because i find soul to be way better at reviewing

442

and digging into the details but i find opus a more pleasant collaborator

443

and significantly better at actually implementing code without getting blocked constantly throughout its work

444

and even now with the rust rewrite of typescript i find

445

myself in a similar pattern where i have opus 5 .5 just going

446

and going and going making the code base work and work well.

447

And then I had Sol come in and do an audit.

448

And this is the funniest part.

449

I remember before I said that I had Sol and Astra working on that TSROS port for effectively months, I told Sol to come in,

450

and it got it from 83 .7 % to 100 % in under a day.

451

I was blown away by that, that it had somehow unblocked the work that Astra and Sol were doing as well as 6 .1 Sol.

452

And I was absolutely blown away by that, that it had taken the work that 5 -6 -Soul, 6 -1 -Soul, and Astra had done over months

453

and got it unblocked where it had been stuck for weeks and finished it.

454

I was much more blown away when I had 6 -1 -Soul take a look at that work and critique it.

455

And what it brought up was that of the 1 .8 million lines of code, 1 .3 million were not being used.

456

The reason was because Opus concluded all of the code from all the other agents was useless slop

457

that had no chance of being recovered.

458

And it chose to rewrite it from scratch itself in another crate.

459

So on one hand, the only reason the code worked was Opus.

460

But on the other hand, the only reason the slop was still around was also Opus.

461

So I had to have this model come in

462

and clean up the mess that other OpenAI models had made because Opus didn't even notice the mess was still there.

463

What I'm trying to say is this model absolutely has a place in your workflows.

464

It could probably even be your default coding model and you wouldn't have too many issues with it.

465

But I still find Opus to be a better collaborator overall.

466

That said, I have almost no reason to use Sonnet anymore because this will effectively take its place.

467

And you bet your butt the moment this model comes out, I'll be going and making adjustments inside of my clod config

468

because I already have it set up so

469

that I can use Sol inside of clod code

470

because I want to make sure Opus knows this is the model to have review its work

471

and investigate the things going on in the code base.

472

This is a damn good model, and I'm really happy to have it.

473

I wish we had something bigger, smarter, and more capable overall.

474

I was really hoping for something to truly dethrone Opus 5 .5 as my daily driver.

475

This isn't it, and I'm not planning on canceling any of my Claude subs as a result of this release, but I am planning on taking a lot more advantage of my Codex subs in my day -to -day work,

476

admittedly in Claude code.

477

This is an awesome release and an unbelievable price for what you're getting, but this does potentially mark the start of the end of the subsidization era, so make sure you're subscribed so that you can be here

478

when I cover all of that

479

and more god I hope this doesn't get me canceled online

480

I have no idea how others feel about this beyond like

481

a handful of early access testers I've talked to I legitimately don't know

482

if you are going to love it or hate it

483

or land somewhere between I will know in a few hours I guess

484

because it's a yeah it's three in the morning I am

485

going to go to bed now this is a tiring one

486

hopefully I did a good job let me know in the comments

487

and until next time peace nerds god I'm so dead All right

📺 同じチャンネル

PirateSoftware is right, this needs to stop

Business EnglishUniversity StudentsWorking Adults

✨ おすすめ動画

66 文

The one OpenAI announcement that can actually make you money...

Business EnglishUniversity StudentsWorking Adults

853 文

Why Coding Agents Keep Making Your Codebase Worse

Business EnglishUniversity StudentsWorking Adults

197 文

Unreal Engine 5.8 - Experimental MCP Server Support Walkthrough | State of Unreal 2026

Business EnglishUniversity StudentsWorking Adults

32 文

このレッスンの語彙とスピーキングのポイント

このC1レベルのスピーキングレッスンは、動画「OpenAI fights back」を教材にしています。繰り返し出てくる語は次のとおりです：model, opus, code, sol, price。この動画には、シャドーイング用の文が487文、単語が6233語あります。音声の長さは30:54です。話す速さは速く、1分あたり約202語です。音のつながりや弱く発音される音が多くなります。英語の頻出3,000語に含まれる単語は81%だけなので、語彙は難しめです。

この動画の重要語彙

動画の中で特に難しい単語15語を、発音と意味つきで紹介します。

単語	発音	意味
sonnet 名詞	/ˈsɒnɪt/	ソネット
token 名詞	/ˈtoʊkən/	トークン
router 名詞	/ˈɹaʊ.tɚ/	ルータ, ルーター
depot 名詞	/ˈdiːpoʊ/	倉庫
merge 動詞	/mɝd͡ʒ/	合わせる, 合併する
frontier 名詞	/ˈfɹʌntɪə/	境界, 国境
audit 名詞	/ˈɔːdɪt/	監査
astrology 名詞	/əˈstɹɒlədʒi/	占星学
rewrite 動詞	/ɹiˈɹaɪt/	書き直す
unbelievable 形容詞	/ˌʌnbɪˈliːvəbl̩/	信じられない
surprisingly 副詞	/səˈpɹaɪzɪŋli/	驚くほど, 意外に
aster 名詞	/ˈæs.tə(ɹ)/	嫁菜, ヨメナ
blender 名詞	/ˈblɛn.dɚ/	ブレンダー, ミキサー
codex 名詞	/ˈkoʊdɛks/	書誌
collaborator 名詞	/kəˈlæbəɹeɪtɚ/	合作者, 協力者

動画に出てくる句動詞

単語	発音	意味
come out 動詞	/ˌkʌm ˈaʊt/	出る
blow away 動詞		吹き飛ぶ
go down 動詞		降りる, 降下する
bring up 動詞		育てる
clean up 動詞		掃除する
come back 動詞		戻る, 帰る
fill out 動詞	/fɪl ˈaʊt/	書き込む, 記入する
grow up 動詞		成長する, 大人に成る

繰り返し練習したい文

動画に出てくる短く完結した文です。日常会話でそのまま使えます。

What am I referring to there?
I didn't save all the prices for these runs.
I'd call that very aggressive pricing.
I'm kind of spoiling the ending here, aren't I?
Said better, but I'll still fix two small gaps.

この動画の文法

話し手がよく使っている文型を、動画の実際の表現とともに紹介します。

文型	動画での表現
現在完了進行形 have/has been + -ing — 以前に始まり今も続いている動作	we've been getting · i've been moving · I've been running
条件文 if + 節, will/would + 動詞 — 条件とその結果	if you do the math, it will net
現在完了形 have/has + 過去分詞 — 過去の出来事が今も関係している	has made · has gone · has not changed
受動態 be + 過去分詞 — 誰がするかより、何が起きるかに焦点を当てる	is called · being paid · was not meant

注意したい発音

話し手はI'm, didn't, I'veなど、短縮形や弱形を115回使っています。聞こえたとおりの短い形で発音しましょう。

「th」の音: anthropic /ænˈθɹɒp.ɪk/, fathom /ˈfæðəm/
「sh」と「zh」の音: benchmark /ˈbɛn(t)ʃmɑːk/, screenshot /ˈskriːn.ʃɑt/, subscription /səbˈskɹɪpʃən/
長い単語（アクセントの位置に注意）: importantly /ɪmˈpɔɹ.tənt.li/, astrology /əˈstɹɒlədʒi/, unbelievable /ˌʌnbɪˈliːvəbl̩/, surprisingly /səˈpɹaɪzɪŋli/, collaborator /kəˈlæbəɹeɪtɚ/

日本語話者が苦手な音：

/r/ と /l/ の区別 — /l/ は舌先を歯茎につけ、/r/ はどこにもつけない: importantly /ɪmˈpɔɹ.tənt.li/, astrology /əˈstɹɒlədʒi/, randomly /ˈɹændm̩li/, surprisingly /səˈpɹaɪzɪŋli/, collaborator /kəˈlæbəɹeɪtɚ/
/f/ — 「フ」ではなく、上の歯と下唇で出す: frontier /ˈfɹʌntɪə/, frustrate /ˈfɹʌsˌtɹeɪt/, fathom /ˈfæðəm/, fidelity /fᵻˈdɛlədi/
子音の連続 — 間に母音を入れない: slop /slɑp/, anthropic /ænˈθɹɒp.ɪk/, frontier /ˈfɹʌntɪə/, importantly /ɪmˈpɔɹ.tənt.li/, astrology /əˈstɹɒlədʒi/

この動画での練習方法

まず声を出さずに動画を最後まで聞き、知らない単語をメモします。
まず0.75倍速で一文ずつシャドーイングし、慣れてきたら通常の速度に戻します。
自分の声を録音して元の音声と比べます。sonnet, token, routerなどの単語に特に注意しましょう。

シャドーイングとは？英語上達に効果的な理由

シャドーイング（Shadowing）は、もともとプロの通訳者養成プログラムで開発された言語学習法で、多言語習得者として知られるDr. Alexander Arguelles によって広く普及されました。方法はシンプルですが非常に効果的：ネイティブスピーカーの英語を聞きながら、1〜2秒の遅延で声に出してすぐに繰り返す——まるで「影（shadow）」のように話者を追いかけます。文法ドリルや受動的なリスニングと異なり、シャドーイングは脳と口の筋肉が同時にリアルタイムで英語を処理・再現することを強制します。研究により、発音精度、抑揚、リズム、連音、リスニング力、そして会話の流暢さが大幅に向上することが確認されています。IELTSスピーキング対策や自然な英語コミュニケーションを目指す方に特におすすめです。

シャドーイングのやり方: ステップ別の完全ガイドを読む →

シャドーイング練習: OpenAI fights back - 動画で英語スピーキングを学ぶ

PirateSoftware is right, this needs to stop

The one OpenAI announcement that can actually make you money...

Why Coding Agents Keep Making Your Codebase Worse

Unreal Engine 5.8 - Experimental MCP Server Support Walkthrough | State of Unreal 2026

Gemini Robotics 2 brings whole body intelligence to robots

The Kurzgesagt Situation is Insane

We Let $200 GPT-Pro Think for 7 Days on Unsolved Math

歐巴馬談 AI：喊放慢的公司「是真的在擔心自己的產品」【AI 大人說】

Introducing GPT-6 Astra for developers

Building Anthropic | A conversation with our co-founders

He’s always wrong

Introducing the Agents API

I Tried PewDiePie's Odysseus AI So You Don't Have To

このレッスンの語彙とスピーキングのポイント

この動画の重要語彙

動画に出てくる句動詞

繰り返し練習したい文

この動画の文法

注意したい発音

この動画での練習方法

シャドーイングとは？英語上達に効果的な理由