Disclaimer: Unless otherwise stated, any opinions expressed below belong solely to the author.

Sadly for young Singaporeans, the median salary for fresh university graduates in full-time employment remained at the same level of S$4,500 in 2025 as in 2024 (the figure excludes bonuses and employer’s CPF), according to the annual Graduate Employment Survey published by the Ministry of Education earlier this month.

What’s more, fewer of them succeeded in finding permanent employment within six months of graduation, at 83.4%, compared to 87.1% in 2024.

Fortunately, however, some good news regarding their future was buried in the data provided by Ministry of Manpower in its labour force report for 2025.

As it turns out, local graduates from the six public, autonomous universities see their salaries surge considerably before they even turn 30, with the median reaching S$6,338 last year, up by 5.7% from S$5,995 in 2024.

They are also way ahead of those in other educational avenues, where monthly incomes are as much as S$2,000 to S$3,000 lower, even after a few years in the labour market.

It confirms what I wrote about last month here, that there are really two Singapores—one inhabited by the tertiary degree holders and the other by everyone else. They live completely different lives due to the gulf in their economic circumstances (happily, over 60% of young Singaporeans complete university education these days).

What’s more, we have to remember that these figures move with time, so we shouldn’t compare the current figure for 25-29 year olds with recent graduates, but rather to what their starting salary was.

Back in 2019, six years ago, it was just S$3,600, so those who graduated at that time could have nearly doubled their money since leaving university.

Another optimistic finding is that these trends continue into your 30s. The median after another decade, for those degree holders aged 35 to 39, is close to S$10,000 per month. And the median for degree holders of all ages is S$9,000 (below is a reminder from my Feb article), putting them way over everybody else.

That’s why, while it may seem that fresh grads are having it a bit more difficult than their predecessors—with stagnating wages and a bit fewer opportunities for quick employment—they really have no reason to worry about the future.

Qualified, university-educated Singaporeans are still the country’s elite.

• Read other articles we’ve written on Singapore’s current affairs here.

Featured Image Credit: National University of Singapore

KRAFTON has just rolled out the BGMI 4.3 update, and it’s easily one of the biggest updates the game has seen so far. This time, the company isn’t just adding a few features; it’s revamping the entire experience alongside partnerships with CSK and KKR, a new Card Collection system, a fresh progression system, and collaborations with Jujutsu Kaisen and Apollo Sportscar. Here’s everything you need to know.

What’s New?

The biggest highlight of the update is the complete redesign of the game’s interface. BGMI now features a brighter, cleaner, and more modern look aimed at improving usability while still keeping the core experience intact. Menus are now less cluttered, with larger tiles and clearer navigation paths. KRAFTON has also added enhanced lobby visuals, dynamic lighting effects, and improved character displays to give the game a more premium feel.

Beyond that, the game is doubling down on India-focused content through collaborations with Chennai Super Kings (CSK) and Kolkata Knight Riders (KKR). The update introduces:

• Interactive elements tied to team identity
• Franchise-themed outfits and vehicle skins
• Dedicated Photobooth POIs in Erangel and Livik

The CSK content goes live on March 21, while KKR arrives on March 25, just in time for the cricket season hype.

New Card Collection

Another big addition is the Card Collection system, which introduces a new way to engage with the game beyond just matches. Players can:

• Earn cards through missions and events
• Trade cards with friends
• Complete themed collections
• Unlock exclusive rewards, including weapon skins

With this update, KRAFTON is clearly positioning BGMI as more than just a battle royale game. The focus is shifting toward making it a live platform that blends gaming with pop culture, collectibles, and community-driven content.

In the last few years, Chinese AI startup MiniMax has become one of the most exciting in the crowded global AI marketplace, carving out a reputation for delivering frontier-level large language models (LLMs) with open source licenses and before that, high-quality AI video generation models (Hailuo).

The release of MiniMax M2.7 today — a new proprietary LLM designed to perform well powering AI agents and as the backend to third-party harnesses and tools like Claude Code, Kilo Code and OpenClaw — marks yet a new milestone: Rather than relying solely on human-led fine-tuning, MiniMax has leveraged M2.7 to build, monitor, and optimize its own reinforcement learning harnesses.

This move toward recursive self-improvement signals a shift in the industry: a future where the models we use are as much the architects of their progress as they are the products of human research. The model is categorized as a reasoning-only text model that delivers intelligence comparable to other leading systems while maintaining significantly higher cost efficiency.

However, with M2.7 being proprietary for now, it is a sign once again that Chinese AI startups — for much of the last year, the standard-bearers in the world of the open source AI frontier, making them appealing for enterprises globally due to low (or no) costs and customization — are shifting strategy and pursuing more proprietary frontier models like U.S. leaders like OpenAI, Google, and Anthropic have been doing for years.

MiniMax becomes the second Chinese startup to release a proprietary cutting-edge LLM in recent months following z.ai with its GLM-5 Turbo, and rumors that Alibaba’s Qwen team is also shifting to proprietary development in the wake of the departure of senior leadership and other researchers.

Technical achievement: The self-evolution loop

The defining characteristic of MiniMax M2.7 is its role in its own creation. According to company documentation, earlier versions of the model were used to build a research agent harness capable of managing data pipelines, training environments, and evaluation infrastructure.

By autonomously triggering log-reading, debugging, and metric analysis, M2.7 handled between 30 percent and 50 percent of its own development workflow.

This is not merely an automation of rote tasks; the model optimized its own programming performance by analyzing failure trajectories and planning code modifications over iterative loops of 100 rounds or more.

“We intentionally trained the model to be better at planning and at clarifying requirements with the user,” explained MiniMax Head of Engineering Skyler Miao on the social network X. “Next step is a more complex user simulator to push this even further.”

This capability extends to complex environments via the MLE Bench Lite, a series of machine learning competitions designed to test autonomous research skills.

In these trials, M2.7 achieved a medal rate of 66.6 percent, a performance level that ties with Google’s new Gemini 3.1 and approaches the current state-of-the-art benchmarks set by Anthropic’s Claude Opus 4.6.

The goal, according to MiniMax, is a transition toward full autonomy in model training and inference architecture without human involvement.

Performance evolution: MiniMax m2.7 vs. m2.5

When compared to its predecessor, M2.5, released in February 2026, the M2.7 model demonstrates significant gains in high-stakes software engineering and professional office tasks.

While M2.5 was celebrated for polyglot code mastery, M2.7 is designed for real-world engineering—tasks requiring causal reasoning within live production systems.

Key performance metrics include:

• Software engineering: M2.7 scored 56.22 percent on the SWE-Pro benchmark, matching the highest levels of global competitors like GPT-5.3-Codex.

• Professional office delivery: In document processing, M2.7 achieved an Elo score of 1495 on GDPval-AA, which the company claims is the highest among open-source-accessible models.

• Hallucination reduction: The model scores plus one on the AA-Omniscience Index, a massive leap from the negative 40 score held by M2.5.

• Hallucination rate: M2.7 achieves a hallucination rate of 34 percent, which is lower than the rates of 46 percent for Claude Sonnet 4.6 and 50 percent for Gemini 3.1 Pro Preview.

• System comprehension: On Terminal Bench 2, the model scored 57.0 percent, demonstrating a deep understanding of complex operational logic rather than simple code generation.

• Skill adherence: On the MM Claw evaluation, which tests 40 complex skills exceeding 2,000 tokens each, M2.7 maintained a 97 percent adherence rate, a substantial improvement over the M2.5 baseline.

• Intelligence parity: The model’s reasoning capabilities are considered equivalent to GLM-5, yet it uses 20 percent fewer output tokens to achieve similar results.

The model’s evolution is further evidenced by its score of 50 on the Artificial Analysis Intelligence Index, representing an 8-point improvement over its predecessor in just one month, and also taking the 8th place overall globally in terms of its overall intelligence across benchmarking tasks in various domains.

Not all independent, third-party benchmarks show improvement for M2.7 over M2.5: On BridgeBench, a set of tasks designed by agentic AI coding startup BridgeMind to test a model’s performance for “vibe coding,” or turning natural language into working code, M2.5 scored 12th place while M2.7 scored 19th place.

Access, pricing, and integration

MiniMax M2.7 is a proprietary model available through the MiniMax API and MiniMax Agent creation platforms. While the core model weights for M2.7 remain closed, the company continues to contribute to the ecosystem through the open-source interactive project OpenRoom.

For direct API integration and via third-party provider OpenRouter, MiniMax M2.7 maintains a cost-leading price point of 0.30 dollars per 1 million input tokens and 1.20 dollars per 1 million output tokens, which is unchanged from the pricing for M2.5. That makes M2.7 one of the most affordable frontier AI models to run in the world — only xAI’s Grok 4.1 Fast is cheaper.

To support different usage scales and modalities, MiniMax offers a structured Token Plan with various subscription tiers. These plans allow users to access models across text, speech, video, image, and music under a single unified quota.

To further drive adoption, MiniMax has launched an Invite and Earn referral program, providing a 10 percent discount to new invitees and a 10 percent rebate voucher to the inviter.

Monthly standard Token Plan pricing: The standard monthly tiers are designed for entry-level developers to heavy regular users.

• Starter: $10 per month for 1,500 requests per 5 hours.

• Plus: $20 per month for 4,500 requests per 5 hours.

• Max: $50 per month for 15,000 requests per 5 hours.

Monthly high-speed Token Plan pricing: For production-scale workloads requiring the M2.7-highspeed variant, the following tiers are available:

• Plus-Highspeed: $40 per month for 4,500 requests per 5 hours.

• Max-Highspeed: $80 per month for 15,000 requests per 5 hours.

• Ultra-High-Speed: $150 per month for 30,000 requests per 5 hours.

Yearly Token Plan pricing: Yearly subscriptions provide significant discounts for long-term commitment:

• Standard Starter: $100 per year (saves 20 dollars).

• Standard Plus: $200 per year (saves 40 dollars).

• Standard Max: $500 per year (saves 100 dollars).

• High-Speed Plus: $400 per year (saves 80 dollars).

• High-Speed Max: $800 per year (saves 160 dollars).

• High-Speed Ultra: $1,500 per year (saves 300 dollars).

One request in these plans is roughly equivalent to one call to MiniMax M2.7, though other models in the suite, such as video or high-definition speech, consume requests at a higher rate.

Official tool integrations

To ensure seamless adoption, MiniMax has provided official documentation for integrating M2.7 into over 11 major developer tools and agent harnesses.

This includes widely used platforms such as Claude Code, Cursor, Trae, and Zed. Other officially supported tools include OpenCode, Kilo Code, Cline, Roo Code, Droid, Grok CLI, and Codex CLI.

Additionally, the model supports the Model Context Protocol, allowing it to natively use tools like Web Search and Understand Image for multimodal reasoning. Developers using the Anthropic SDK can easily integrate M2.7 by modifying the ANTHROPIC_BASE_URL to point to the MiniMax endpoint.

When using MiniMax as a provider in tools like OpenClaw, image understanding capabilities are automatically configured via the model’s VLM API endpoint, requiring no extra setup from the user.

With its deep bench of integrations and its pioneering approach to recursive self-evolution, MiniMax M2.7 represents a significant step toward an AI-native future where models are as involved in their own progress as the humans who guide them.

Strategic implications for enterprise decision-makers

Technical decision-makers should interpret the M2.7 release as evidence that agentic AI has moved from theoretical prototyping to production-ready utility.

The model’s ability to reduce recovery time for live production incidents to under three minutes by autonomously correlating monitoring metrics with code repositories suggests a paradigm shift for SRE and DevOps teams.

Enterprises currently facing pressure to adopt AI-driven efficiencies must decide whether they are content with AI as a sophisticated assistant or if they are ready to integrate native agent teams capable of end-to-end full project delivery.

From a financial perspective, M2.7 represents a significant breakthrough in cost efficiency for high-level reasoning. Analysis indicates that M2.7 costs less than one-third as much to run as GLM-5 at equivalent intelligence levels.

For example, running a standard intelligence index cost 176 dollars on M2.7 compared to 547 dollars for GLM-5 and 371 dollars for Kimi K2.5. This aggressive pricing strategy places M2.7 on the Pareto frontier of the intelligence vs. cost chart, offering enterprise-level reasoning at a fraction of the market rate.

The current market is saturated with high-performance models, many of which still hold slight edges in general reasoning scores. But the specific optimization of M2.7 for Office Suite fidelity in Excel, PPT, and Word and its high performance in the GDPval-AA benchmark make it a primary candidate for organizations focused on professional document workflows and financial modeling.

Decision-makers must weigh the benefits of a general-purpose frontier model against a specialized engine like M2.7, which is built to interact with complex internal scaffolds and toolsets.

Ultimately, the fact that it is fielded by a Chinese company (headquartered in Shanghai) and subject to that country’s laws in addition to the user’s country, and is not available for offline or local usage yet, may make it a tough sell for enterprises operating in the U.S. and the West — especially those in highly-regulated or government-facing industries.

Nonetheless, the shift toward self-evolving models suggests that the ROI of AI investment will increasingly be tied to the recursive gains of the system itself.

Organizations that adopt models capable of improving their own harnesses may find themselves on a faster iteration curve than those relying on static, human-only refinement. With MiniMax’s aggressive integration into the modern developer stack, the barrier to testing these autonomous workflows has dropped significantly, placing pressure on competitors to deliver similar native agent capabilities.

Marshall has just introduced a smaller variant of its largest party speaker for $500 less. The new Marshall Bromley 450 ($799) follows the Bromley 750 ($1,299) from last year, while keeping the same design, light show and retro style in smaller footprint.

The Bromley 450 also maintains very similar specifications including True Stereophonic 360-degree sound, over 40 hours of battery life, and integrated stage lighting that leans hard into vintage concert vibes without feeling like a gimmick. Wrapped in water-based PU leather with reinforced corners and a metal grille, it’s built to survive actual parties—not just look good in marketing shots.

Designed for both indoor and outdoor use, the Bromley 450 delivers wide, room-filling sound with a portable footprint and built-in handle that makes it easy to haul from backyard to beach. Add in dual combo inputs for microphones or instruments, and it’s clear Marshall isn’t just chasing the Bluetooth crowd, it’s going after anyone who thinks a speaker should do more than just sit there.

“Last year, we launched our first-ever party speaker, and we’re excited about growing that family of Marshall speakers already this Spring,” said Hanna Wallner, Product Manager at Marshall Group. “Bromley 450 is the natural continuation and smaller sibling of Bromley 750. This speaker is smaller and more affordable yet still packed with impressive features, including sound that hits every corner, a stage light-inspired light show, and our unique Marshall design.”

Marshall Bromley 450 Features

True Stereophonic 360° Sound: Designed for both indoor and outdoor use, the Bromley 450 uses multi-directional driver placement to deliver consistent sound coverage around the speaker. This helps maintain a balanced listening experience without requiring a fixed listening position.

40+ Hours of Portable Playtime: The Bromley 450 can operate on AC power or its built-in rechargeable battery. On battery, it provides over 40 hours of playback, depending on volume levels and usage. An optional spare battery can be used to extend listening time.

Integrated Stage Lights: The speaker includes built-in lighting with three selectable presets. One provides ambient illumination, while the others respond dynamically to the music to add visual elements to playback.

Mic and Instrument Inputs: The Bromley 450 includes inputs for microphones and instruments, enabling use for karaoke or live performance. Additional connectivity includes RCA and USB-C inputs for external audio sources.

Bluetooth Connectivity: The Bromley 450 supports wireless Bluetooth playback from smartphones, tablets, and computers. It is also compatible with Auracast, allowing it to pair with other supported speakers through the Marshall Bluetooth app for expanded coverage.

Dirt and Splash Resistance: Rated IP55, the Bromley 450 is protected against dust and low-pressure water exposure. This makes it suitable for outdoor use in environments such as backyards, parks, or festivals, where light splashes or debris may be present.

Built-in Handle: The Bromley 450 includes an integrated handle for easier transport. Its size and design make it practical to carry between locations or load into a vehicle for outdoor use.

Replaceable Battery: The speaker uses a replaceable LFP battery with adaptive charging designed to help maintain long-term battery health. Replacement batteries are available through Marshall.

Recycled Materials: The Bromley 450 is constructed with 15% recycled material by weight, reducing the environmental impact associated with raw material sourcing and processing.

“With Bromley 450, our goal was to take everything we loved about the Bromley 750 and bring it into a more compact form,” said Malcolm Kennedy, Director of Audio & Acoustics at Marshall Group. “It delivers the same signature sound: fast, powerful bass, clean mids, and detailed highs. Thanks to our unique 360-degree True Stereophonic design, it can always be the center of the party, preserving the depth and atmosphere of your favorite music.”

Comparison

The Bottom Line 

The Marshall Bromley 450 stands out by combining PA-style functionality with consumer-friendly design. True Stereophonic 360-degree sound, long battery life, replaceable power, and support for microphones and instruments make it more flexible than a typical Bluetooth speaker. Auracast compatibility also gives it room to scale beyond a single unit, which is becoming more relevant as multi-speaker ecosystems evolve.

What it does not emphasize is app-driven smart features, voice assistants, or advanced streaming platform integration. This is not a Wi-Fi multi-room speaker or a high-resolution streaming hub. It is focused on local playback, physical inputs, and straightforward Bluetooth use.

It’s designed for users who want a portable, durable speaker that can handle social settings, small events, or outdoor gatherings without needing additional gear. It will appeal to those who value physical connectivity and long battery life over smart features, and who want something that can function as both a speaker and a simple event system for under $800.

Price & Availability

For more information, visit Marshall’s party speaker page.

Related Reading:

The pair are considered originators in their field, which blends physics and computer science in treating quantum mechanical phenomena as resources for processing and transmitting information.

This year’s Turing Award has gone to an American physicist and a Canadian computer scientist for their foundational collaborative work in the field of quantum information science.

Charles H Bennett and Gilles Brassard received the annual ACM (Association for Computing Machinery) award “for their essential role in establishing the foundations of quantum information science and transforming secure communication and computing”, said the body.

The pair’s pioneering work in quantum cryptography and quantum teleportation is recognised for having redefined secure communication and computing, according to the ACM.

The award, often referred to as the ‘Nobel Prize in Computing’, is named after Alan Turing, who articulated the mathematical foundations of computing. The winner receives a $1m prize in recognition of their major contributions of lasting importance to computing.

Bennett and Brassard are considered originators in their field, which blends physics and computer science in treating quantum mechanical phenomena as resources for processing and transmitting information.

In 1984, the pair introduced the first practical protocol for quantum cryptography, now known as BB84, by demonstrating that two parties could establish a secret encryption key with security guaranteed by the laws of physics.

This established a fundamental property of quantum information: it cannot be copied or measured without disturbance, and any attempt at ‘eavesdropping’ leaves detectable traces before any information can be compromised.

Prior to this breakthrough, the consensus around secure communications held that mathematical and computational encryption barriers were the foundation of information secrecy.

“Bennett and Brassard fundamentally changed our understanding of information itself,” said ACM president Yannis Ioannidis. “Their insights expanded the boundaries of computing and set in motion decades of discovery across disciplines. The global momentum behind quantum technologies today underscores the enduring importance of their contributions.”

Variants of BB84 have already been implemented in operational quantum communication networks around the world, using both landlines via fibre and free space communication through satellites, according to the ACM, which also noted that progress in this arena could represent one pathway for achieving secure digital communications in the coming decades.

“Charles Bennett and Gilles Brassard’s visionary insights laid the groundwork for one of the most exciting frontiers in science and technology,” said Jeff Dean, a chief scientist at Google DeepMind and Google Research. “Their work continues to influence both fundamental research and real-world innovation.” Google gives financial support to the annual award.

Bennett and Brassard’s other work of note includes contributions in quantum teleportation and entanglement, which are significant to the application of quantum networking.

In Europe, France’s Pasqal and Finland’s IQM are significant players in the quantum computing sector.

In Ireland, interest in the quantum computing sector features at both private and public levels.

Last year, Andrew Barto and Richard Sutton won the Turing award for developing the foundations of reinforcement learning, which is key to AI. Previous winners include theoretical computer scientist Avi Wigderson, AI leader Geoffrey Hinton and Lisp programming inventor John McCarthy.

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.

There are tons of colors, covers, and other accessories you can get for it, too. I recommend picking out a protective cover since I did ding up my soft cover quite a bit, and that was just from being used around the house with an occasional trip in a bag. (The Techo also has a nice faux leather cover option, if that’s more your style than a protective cover.) There are little quotes in the corners, but I got a Japanese version instead of an English one, so I can’t actually read mine. But honestly, I’m glad I did; I prefer the Japanese lettering to distracting English words on the corners of each page.

The Kokuyo Jibun Techo is another Japanese planner (as indicated by the Japanese word “techo,” meaning “notebook” or “planner”), but this one is a travel-notebook-style design, making it more customizable with a cover and multiple books that fit inside it. I’ve started using the Jibun Techo First Kit, which comes with three books that fit into the included cover: Diary (labeled simply with the year 2026), Life, and Idea. The Diary is in the center and is the true planner, with monthly spreads and weekly spreads for the whole year, plus other fun pages. The Life book goes in the front and features lots of specific prompts, like the 100 Wishes List and places to track specific information, like passwords. Finally, the Idea book, which goes in the back, has just grid paper, so you can write out whatever you’d like in whatever style.

It’s a really fun planner thanks to all the interesting pages it comes with, and how much space you have to work with. I love turning the project planner pages into a habit tracker, for example, but I also love using the prompt pages like 100 Wishes as they’re designed. Similar to the Hobonichi above, the Jibun Techo uses a super-thin, super-soft paper that makes it possible to include all these pages and books without making the whole planner super-thick. It’s also smooth and satisfying to write on, thanks to that nice paper. It’s a bigger investment than the Roterunner and I don’t use every single page so far, but it’s still a fantastic planner if you want an option with tons of different pages you can use for planning and reflecting on everything in your life.

Honorable Mentions

Like I said, there’s a huge world of planners out there. Here are a few that the Reviews team and I at WIRED have tested and enjoyed.

The ability to connect to multiple high-resolution displays even means it can comfortably run a full workstation, so long as you get a USB hub or Thunderbolt dock for more ports. That’s the power of those two Thunderbolt 4 ports onboard that you’ll find on the left side of the MacBook Air. My only complaint is that they’re all on the left, meaning you can’t charge the device from the right side. Meanwhile, storage speed is on par with the M5 MacBook Pro (and around six times faster than the MacBook Neo), with an average read/write speed of 6,740 megabytes per second. That is an exponential jump in speed over all the previous MacBook Airs, and that accelerates everything from opening applications to transferring large files. It’s one of the reasons (along with the limitation of 16 GB of RAM) that people who spend eight hours or more working on a laptop every day will want the MacBook Air over the MacBook Neo.

Photograph: Luke Larsen

The Windows side is also offering some serious competition. I still love the Surface Laptop as an alternative, though we’re waiting for the next-generation model to come. The Dell XPS 13, currently only $850, is a closer size to the 13-inch MacBook Air and uses the same Qualcomm Snapdragon X Plus chips as the Surface Laptop. Unfortunately, it doesn’t have as high-resolution and bright a screen. The extremely lightweight Asus Zenbook A14 can’t match the MacBook Air’s screen either, but is often on sale for hundreds of dollars less.

Lastly, there’s the 2025 M4 MacBook Air to consider. If you never use heavier applications, you won’t benefit much from the difference in the M5’s performance. The storage options are also slightly different from last year’s—Apple removed the 256-GB model from the lineup, and the M5 starts at 512 GB. You might look at the prices on Amazon and be tempted to pocket some cash by getting the M4. But, barring a sale of some kind, the 512-GB M5 MacBook Air will only cost $50 more than the 512-GB M4.

Despite the compelling options presented by both the MacBook Neo and an assortment of different Windows laptops, the M5 MacBook Air is still the laptop most people should buy. It’s well rounded, surprisingly powerful, high-end, and will last you many, many years to come.