OpenAI responds to DeepSeek competition with detailed reasoning traces for o3-mini - Related to completely, this, pc, competition, accidentally

OpenAI responds to DeepSeek competition with detailed reasoning traces for o3-mini

OpenAI is now showing more details of the reasoning process of o3-mini, its latest reasoning model. The change was revealed on OpenAI’s X account and comes as the AI lab is under increased pressure by DeepSeek-R1, a rival open model that fully displays its reasoning tokens.

Models like o3 and R1 undergo a lengthy “chain of thought” (CoT) process in which they generate extra tokens to break down the problem, reason about and test different answers and reach a final solution. Previously, OpenAI’s reasoning models hid their chain of thought and only produced a high-level overview of reasoning steps. This made it difficult for clients and developers to understand the model’s reasoning logic and change their instructions and prompts to steer it in the right direction.

OpenAI considered chain of thought a competitive advantage and hid it to prevent rivals from copying to train their models. But with R1 and other open models showing their full reasoning trace, the lack of transparency becomes a disadvantage for OpenAI.

The new version of o3-mini displays a more detailed version of CoT. Although we still don’t see the raw tokens, it provides much more clarity on the reasoning process.

In our previous experiments on o1 and R1, we found that o1 was slightly improved at solving data analysis and reasoning problems. However, one of the key limitations was that there was no way to figure out why the model made mistakes — and it often made mistakes when faced with messy real-world data obtained from the web. On the other hand, R1’s chain of thought enabled us to troubleshoot the problems and change our prompts to improve reasoning.

For example, in one of our experiments, both models failed to provide the correct answer. But thanks to R1’s detailed chain of thought, we were able to find out that the problem was not with the model itself but with the retrieval stage that gathered information from the web. In other experiments, R1’s chain of thought was able to provide us with hints when it failed to parse the information we provided it, while o1 only gave us a very rough overview of how it was formulating its response.

We tested the new o3-mini model on a variant of a previous experiment we ran with o1. We provided the model with a text file containing prices of various stocks from January 2024 through January 2025. The file was noisy and unformatted, a mixture of plain text and HTML elements. We then asked the model to calculate the value of a portfolio that invested $140 in the Magnificent 7 stocks on the first day of each month from January 2024 to January 2025, distributed evenly across all stocks (we used the term “Mag 7” in the prompt to make it a bit more challenging).

o3-mini’s CoT was really helpful this time. First, the model reasoned about what the Mag 7 was, filtered the data to only keep the relevant stocks (to make the problem challenging, we added a few non–Mag 7 stocks to the data), calculated the monthly amount to invest in each stock, and made the final calculations to provide the correct answer (the portfolio would be worth around $2,200 at the latest time registered in the data we provided to the model).

It will take a lot more testing to see the limits of the new chain of thought, since OpenAI is still hiding a lot of details. But in our vibe checks, it seems that the new format is much more useful.

When DeepSeek-R1 was released, it had three clear advantages over OpenAI’s reasoning models: It was open, cheap and transparent.

Since then, OpenAI has managed to shorten the gap. While o1 costs $60 per million output tokens, o3-mini costs just $[website], while outperforming o1 on many reasoning benchmarks. R1 costs around $7 and $8 per million tokens on [website] providers. (DeepSeek offers R1 at $[website] per million tokens on its own servers, but many organizations will not be able to use it because it is hosted in China.).

With the new change to the CoT output, OpenAI has managed to somewhat work around the transparency problem.

It remains to be seen what OpenAI will do about open sourcing its models. Since its release, R1 has already been adapted, forked and hosted by many different labs and companies potentially making it the preferred reasoning model for enterprises. OpenAI CEO Sam Altman not long ago admitted that he was “on the wrong side of history” in open source debate. We’ll have to see how this realization will manifest itself in OpenAI’s future releases.

The Peak XV-backed B2B marketplace has finalised six bankers to helm its upcoming IPO.

The startup received the board’s approval for the fundraise, wh...

The Dell G16, a gaming laptop that provides amazing value for money, is currently even more affordable right now when you buy from Dell. This particul...

This adorable Noctua cooler completely transformed my gaming PC

Table of Contents Table of Contents Meet the 92mm beast Small PC, big performance.

About a year ago, I challenged myself. I wanted to know how small I could make a PC while packing in the highest-end gaming hardware money can buy, and that’s sent me on a bit of a journey.

I’ve made tweaks here and there, swapped out hardware, and endlessly fiddled with my fan curves. But finally, after many months and lot of money down the drain, it feels like my PC has reached its final form. And it all happened because I invested about $60 in a tiny Noctua cooler that’s completely changed my relationship with my small form factor (SFF) PC.

OK, enough preamble. This mystery Noctua cooler is the NH-L9x65. It’s available in both Noctua’s [website] color option, as well as silver with one of Noctua’s iconic beige and brown fans. I chose the latter because I love the look — we exist — but it’s what this cooler has going on under the hood that stands out.

It’s a low-profile cooler, clocking in at just 65mm tall, and that’s a requirement given that my PC is built inside of a Fractal Terra mini-ITX case — the best mini-ITX case you can buy, thank you very much. I wasn’t able to use a cooler this tall previously, as I tried cramming in an RTX 4090 and was only left with about 55mm of clearance for a CPU cooler. But a lot has changed since then.

Notably, Nvidia released the RTX 5090. This is a proper two-slot graphics card — at least for the Founder’s Edition model — freeing up a ton of extra space for a CPU cooler. Previously, I was using an ID-Cooling IS-55, which was just barely enough to keep the Ryzen 7 9700X in my PC cool. Jumping from the RTX 4090 to the RTX 5090 opened up more room, allowing me to get a taller cooler, and in turn, pack in a more powerful CPU (more on that in a moment).

The NH-L9x65 is a taller cooler, but it’s not a bigger cooler, and that’s why I love it so much. Instead of a standard 120mm fan, the NH-L9x65 uses one of Noctua’s slim 92mm NF-A9x14 fans. If you’ve worked with mini-ITX motherboards before, you know that’s a big deal. With Noctua’s cooler, I don’t have to fight with my case or the heatsinks built onto the motherboard, and that made properly installing it a breeze.

Due to how small mini-ITX motherboards are, you’re given a limited footprint for a cooler. In the case of something like the IS-55 I was using previously, the cooler actually had a portion of the heatsink clipped out to make room either for the RAM or the motherboard’s heatsinks. Not a problem on the L9x65. The cooler doesn’t protrude too far beyond the CPU socket, making the installation far easier.

The difference-maker is the performance, however. The NH-L9x65, despite its petite size, is one hell of a CPU cooler, and it’s completely changed my SFF rig.

I settled for the Ryzen 7 9700X in the previous iteration of my PC. I intended to use the Ryzen 7 9800X3D, but given the space and cooler constraints, I decided to ditch the 3D V-Cache chip for a standard 8-core offering. The NH-L9x65 made AMD’s best gaming CPU a possibility, though, and the performance has been fantastic.

Here’s a taste. Below, you can see my temperatures with Noctua’s cooler, and the results are great. For a small form factor PC with a high-end gaming CPU known to run hot, I’ll take idle temperatures under 50 degrees Celsius. And although the temperature ramped up with an all-core Cinebench R24 load, it’s still well within safe operating temperatures. The chip also maintained its maximum boost clock of [website] across all cores during the run, which is mighty impressive.

Idle (30 minutes on Windows desktop) [website] degrees Celsius Daily use (10 Chrome tabs, Discord, and Steam) [website] degrees Celsius Cinebench R24 all-core [website] degrees Celsius.

These might seem a little warm if you’re accustomed to a full-sized desktop, but for a case as tiny as the Fractal Terra — and packing some power-hungry hardware — this is exceptional performance. I’ve seen the CPU break 90 degrees before, but only for a brief moment, and while I’m playing games, it rarely goes above 70 degrees.

Temperatures are great, but given the size of the NH-L9x65, I assumed it would make trade-offs in noise. This is a 92mm fan, after all, and the smaller you go on the fan, the more loud and annoying the whine is when the fan ramps up. But no. The NH-L9x65 is cool, but it’s also remarkably quiet given the hardware it has to cool.

0% fan speed [website] decibels 50% fan speed [website] decibels 100% fan speed [website] decibels.

Decibel measurements need a bit of context. On a noise scale, 40 decibels is considered the average noise inside a house, so my measurement at 0% fan speed represents the ambient noise in my office. Even ramping up to 50% fan speed, the noise is barely audible over the sound of a normal room. For my use, the fan speed hovers between 50% and 70% while I’m working with a dozen or so Chrome tabs open, as well as Discord and Steam running in the background.

In practical use, that means I barely hear my PC while I’m working. That’s great.

When you push the fan hard — and it will get pushed hard while playing games — the noise ramps up quite a bit. Still, [website] decibels isn’t bad. On the decibel scale, this would be just below the noise of a normal conversation (no shouting) and just above the idle hum of a refrigerator. Basically, it doesn’t sound like a jet engine, even when the fan speed is spinning at its maximum RPM and the cores are fully loaded. I’ll take that any day of the week.

In my previous build, I was constantly fighting noise and temperatures, regardless of how much I tweaked my fan curve, and while using a weaker CPU. Now, I’m able to get advanced performance without worrying that my PC is going to take flight with how fast the fan is spinning. And that largely comes down to spending $60 on a new CPU cooler.

The deal is expected to complete within the period of 90 days from the date of approval by the Zaggle’s board.

The financial terms of the deal remain ...

The Dell G16, a gaming laptop that provides amazing value for money, is currently even more affordable right now when you buy from Dell. This particul...

Spigen just accidentally leaked iPhone SE 4 renders

The iPhone SE 4 has been a highly-anticipated handset for a while now, and we expect it to drop sometime next week. We just got another good look at it, courtesy of case manufacturer Spigen. The firm uploaded images of its case to its website, along with an iPhone inside the case. The website says it’s an iPhone SE (3rd gen), but one look at the images exhibits that isn’t the case.

Of course, we already had a solid idea of what the iPhone SE 4 would look like. The renders don’t really come with any surprises; in many ways, the iPhone SE 4 looks like the iPhone 14, complete with the notch at the top. The case renders also show a single camera on the rear of the phone. One interesting change is the Alert Slider — the button on the side of your iPhone that enables/disables vibration — seems to have been replaced with an Action Button instead.

The renders provide even more confirmation that the iPhone SE 4 will come with a [website] display. This handset promises to pack a lot more power into its small frame than its predecessors did, making it a budget phone with non-budget power.

Apple isn’t holding an event next week, but insider Mark Gurman has suggested it could release on February 12 or 13, and February 18 at the latest. Rather than an event, Gurman says interested fans should look for a press release and a video on Apple’s YouTube channel. It goes without saying that you can also find any news related to the iPhone SE 4’s launch here at Digital Trends.

Please enable Javascript to view this content.

There’s still no official word yet on pricing, but those in the know say the iPhone SE 4 could be a bit more expensive than the current model, coming in closer to $499 than $429.

MobiKwik posted a net loss of INR [website] Cr in Q3 FY25 against a profit of INR [website] Cr, impacted by lower financial services revenue and higher lending...

The New York Times has plenty of word games on its roster today — with Wordle, Connections, Strands, and the Mini Crossword, there’s something for eve...