What to be thankful for in AI in 2025
Hello, dear readers. Happy belated Thanksgiving and Black Friday!
This year felt like I was living in a permanent DevDay. Every week a lab comes out with a new model, a new agent framework, or a new “this changes everything” demo. It’s overwhelming. But it’s also the first year that I feel like AI is finally diversifying – not just one or two frontier models in the cloud, but a whole ecosystem: open and closed, big and small, Western and Chinese, cloud and on-premise.
So for this Thanksgiving, here’s what I’m truly grateful for in AI in 2025: the releases that feel like they’ll matter 12 to 24 months from now, and not just during this week’s hype cycle.
1. OpenAI continued to ship strongly: GPT-5, GPT-5.1, Atlas, Sora 2 and open weights
As the company that undeniably ushered in the “generative AI” era in late 2022 with its viral hit product ChatGPT, OpenAI had arguably one of the toughest tasks of any AI company in 2025: continuing its growth trajectory even as well-funded competitors like Google with its Gemini models and other startups like Anthropic brought their own highly competitive offerings to market.
Luckily, OpenAI rose to the challenge and then some. The main activity was GPT-5, unveiled in August as the next groundbreaking reasoning model, followed in November by GPT-5.1 with new Instant and Thinking variants that dynamically adjust how much ‘thinking time’ they spend per task.
In practice, the GPT-5 launch was bumpy – VentureBeat documented early math and coding errors and a cooler-than-expected response from the community in “OpenAI’s GPT-5 rollout isn’t going smoothly”, but this was quickly corrected based on user feedback and as a daily user of this model, I’m personally happy and impressed with it.
At the same time, the companies that actually use the models are reporting solid profits. ZenDesk GlobalFor example, according to GPT-5 agents, they now resolve more than half of customer tickets, with some customers seeing an 80 to 90% resolution rate. That’s the quiet story: these models may not always impress the chattering classes at X, but they are starting to move real KPIs.
On the tooling side, OpenAI finally gave developers a serious AI engineer with GPT-5.1-Codex-Max, a new coding model that can run long, agentic workflows and is already the standard in OpenAI’s Codex environment. VentureBeat covered it in detail in “OpenAI Introduces GPT-5.1-Codex-Max Encryption Model and It Has Already Completed a 24-Hour Task Internally.”
Then there’s ChatGPT Atlas, a full browser that has ChatGPT built into Chrome itself: sidebar summaries, on-page analytics, and search tightly integrated into regular browsing. It’s the clearest sign yet that “assistant” and “browser” are on a collision course.
On the media side, Sora 2 turned the original Sora video demo into a full video and audio model with better physics, synchronized sound and dialogue, and more control over style and shot structure, plus a dedicated Sora app with a full-fledged social networking component, allowing any user create their own TV network in their pocket.
Finally – and perhaps most symbolically – OpenAI has released gpt-oss-120B and gpt-oss-20B, open-weight MoE reasoning models under an Apache 2.0-like license. Whatever you think of its quality (and early open source adopters were vocal about their complaints), this is the first time since GPT-2 that OpenAI has put serious weight into the public commons.
2. The Chinese open source wave is going mainstream
If 2023-2024 was about Lama and Mistral, 2025 belongs to China’s open ecosystem.
This is evident from a study by MIT and Hugging Face China is now slightly ahead of the US in global downloads of open modelslargely thanks to DeepSeek and Alibaba’s Qwen family.
Highlights:
-
DeepSeek-R1 was launched in January as an open-source reasoning model to rival OpenAI’s o1, with MIT-licensed weights and a family of distilled smaller models. VentureBeat has been following the story from release, the cybersecurity impact, and the performance-tuned R1 variants.
-
Kimi K2 Thinks from Moonshot, a ‘thinking’ open source model that reasons step by step with tools, largely in the o1/R1 form, and is positioned as the world’s best open reasoning model to date.
-
Z.ai ships GLM-4.5 and GLM-4.5-Air as “agentic” models, open-sourced base, and hybrid reasoning variants on GitHub.
-
from Baidu ERNIE 4.5 family came as a fully open-source, multi-modal MoE suite under Apache 2.0, including a 0.3B compact model and visual ‘thinking’ variants focused on graphs, STEM and tool usage.
-
Alibabas Qwen3 line – including Qwen3-Coder, large reasoning models and the Qwen3-VL series released in the summer and fall months of 2025 – continues to raise the bar for open weights in coding, translation and multimodal reasoning, leading me to declare this past summer as ‘
Qwen’s summer.”
VentureBeat has been tracking these shifts, including Chinese math and reasoning models like Light-R1-32B and Weibo’s tiny VibeThinker-1.5B, which beat DeepSeek’s baselines on a small training budget.
If you’re interested in open ecosystems or on-site options, this is the year that China’s open-weight scene has gone from being a curiosity to a serious alternative.
3. Small and local models are coming of age
Another thing I’m grateful for: the time has finally come Good small models, not just toys.
Liquid AI has been promoting its Liquid Foundation Models (LFM2) and LFM2-VL vision language variants in 2025, which were designed from day one for low-latency device-aware deployments: edge boxes, robots and limited servers, not just giant clusters. The newer LFM2-VL-3B focuses on embedded robotics and industrial autonomy, with demos planned at ROSCon.
On the big-tech side, Google’s Gemma 3 line made a strong case that “small” can still be capable. Gemma 3 covers parameters from 270M to 27B, all with open weights and multimodal support in the larger variants.
The highlight is the Gemma 3 270M, a compact model purpose-built for fine-tuning and structured text tasks (think custom formatters, routers, and watchdogs) that are covered both in Google’s developer blog and in community discussions in local LLM circles.
These models may never be popular on
4. Meta + Midjourney: aesthetics as service
One of the stranger twists this year: Meta teamed up with Midjourney instead of simply trying to defeat it.
In August, Meta announced a deal to license Midjourney’s “aesthetic technology” – its image and video generation stack – and integrate it into Meta’s future models and products, from Facebook and Instagram feeds to Meta AI features.
VentureBeat described the partnership in “Meta is working with Midjourney and will license its technology for future models and products,” which begged the obvious question: does this slow down or change Midjourney’s own API roadmap? I’m still waiting for an answer, but unfortunately the announced plans for an API release have yet to materialize, suggesting this is the case.
For creators and brands, however, the immediate implication is simple: mid-journey visuals will appear in regular social tools instead of being locked away in a Discord bot. That could normalize higher-quality AI art for a much wider audience – and force rivals like OpenAI, Google and Black Forest Labs to keep raising the bar.
5. Google’s Gemini 3 and Nano Banana Pro
Google tried to answer GPT-5 with Gemini 3, billed as the most capable model yet, with better reasoning, coding and multimodal understanding, plus a new Deep Think mode for slow, difficult problems.
VentureBeat’s coverage, “Google unveils Gemini 3 and claims leadership in math, science, multimodal and agentic AI,” described it as an immediate opportunity for groundbreaking benchmarks and agentic workflows.
But the surprise hit is Nano Banana Pro (Gemini 3 Pro Image), Google’s new flagship image generator. It specializes in infographics, diagrams, multi-topic scenes, and multilingual text that actually appears readable in 2K and 4K resolutions.
In the world of business AI—where graphs, product diagrams, and “visually explain this system” images are more important than fantasy dragons—that’s a big deal.
6. Wildcards I’m keeping an eye on
A few more releases I’m grateful for, even if they don’t fit neatly into one bucket:
-
Flux.2 from Black Forest Labs image models, which were launched earlier this week with the ambition to challenge both Nano Banana Pro and Midjourney in terms of quality and control. VentureBeat dug into the details in “Black Forest Labs Launches Flux.2 AI Image Models to Challenge Nano Banana Pro and Midjourney.”
-
Claude Opus 4.5 by Anthropica new flagship that focuses on cheaper, more capable coding and task execution over a long horizon, covered in “Claude Opus 4.5 from Anthropic is here: cheaper AI, infinite chats and coding skills that beat humans.”
-
A steady drumbeat of open math/reasoning models – from Light-R1 to VibeThinker and others – showing that you don’t need $100 million training runs to move the needle.
Final Thoughts (for now)
If 2024 was the year of “one big model in the cloud,” then 2025 is the year the map exploded: multiple borders at the top, China leading the way in open models, small and efficient systems maturing quickly, and creative ecosystems like Midjourney entering the big tech stacks.
I’m grateful not only for any model, but for the fact that we have it now options – closed and open, local and hosted, reasoning first and media first. For journalists, builders and companies, that diversity is the real story of 2025.
Happy holidays and all the best to you and your loved ones!
