12 AI Highlights: From 2T Models to the Agentic Coding Wave

Bài viết do Ban Biên Tập Marketing365 thực hiện, biên tập theo Chính sách biên tập của Marketing365. Cập nhật lần cuối 21/07/2026.

Nội dung

Claude Code is “upgraded” into a multi-task software team
Elon Musk says SpaceXAI’s 2T-parameter model is nearing the end of its first training run
A dense wave of AI releases is expected in August
The free path to becoming an AI engineer is still wide open
Moonshot’s Kimi K3 makes a splash on the AI coding leaderboard
Grok 4.5 appears on Cognition’s new FrontierCode leaderboard
The Ukraine personnel shake-up story is not the core AI angle of this roundup
Grok 4.5 is praised as very strong, and expectations are centered on the 2T model
Anthropic brings Fable 5 into subscriptions: a sign that price pressure is rising
Grok Build aims for an “AI working alongside you on your laptop” experience
DeepSeek V4 is preparing for GA, with price seen as a major advantage
The AI race between the US and China is entering a “dual race”
A perspective for the Vietnamese market
References

The AI market continues to accelerate, with notable signals across foundation models, agent tools, and the coding benchmark race. For Vietnamese marketers, this is not just a technology story: it directly affects content production, workflow automation, data analysis, and the choice of tools inside the enterprise.

In this roundup, Marketing365 compiles 12 updates that the international community is discussing most, helping you see the competitive landscape of AI more clearly and draw out what can be applied in practice for marketing, product, and operations teams.

Key points:

The AI race is shifting strongly toward “agentic workflows” and real-world coding, no longer just theoretical benchmarks.
Many large models are being mentioned with 2T–2.8T parameter ambitions, showing pressure to compete on both scale and performance.
Price, access, and real-world usability are becoming more important competitive advantages than brand reputation alone.
AI learners can take advantage of an increasingly rich ecosystem of free resources to keep up with the trend.

Claude Code is “upgraded” into a multi-task software team

An open-source repo is drawing attention for turning Claude Code into a workflow that feels close to a miniature “software team”: planning, assigning work, tracking progress, and shipping features through to completion. According to the source description, the CCPM project uses a 5-step process, creates a PRD, converts it into a technical epic, breaks it into up to 10 tasks, and then syncs directly into GitHub Issues.

What stands out is that this approach does not rely on “continuous prompting” but instead organizes AI as a system: multiple agents can run at the same time for the database, backend, API, frontend, and testing; it also uses git worktree to avoid file conflicts. For businesses, this is a clear signal that the real value of AI is shifting from answering questions to coordinating work.

Source: @gippp69 / @leopardracer

Elon Musk says SpaceXAI’s 2T-parameter model is nearing the end of its first training run

Information circulating on X says Elon Musk stated that SpaceXAI’s new 2 trillion-parameter model will complete initial training next week. According to the quoted content, the model is expected to outperform the 1.5T version in every respect, potentially surpass Kimi K3 while still maintaining speed and token efficiency close to Grok 4.5.

If this claim proves true, it would be a notable step because it shows the AI race is no longer just about scaling parameter counts, but also about preserving inference performance. For marketers, the lesson is that “massive” models will keep arriving, but the ability to integrate them into real workflows is what will determine who actually uses them regularly.

Source: @mark_k / @elonmusk

A dense wave of AI releases is expected in August

A list widely shared on social media suggests August could be a packed month for AI launches: Claude Opus 5, GPT-6, Claude Fable 5.1, Grok 2T, GLM-5.3, Gemini 3.5 Pro, Gemini 3.6 Flash, DeepSeek V4 GA, Qwen 3.8, MiniMax M3.1, and Composer 3. While most of these are still speculation or on a “watchlist,” it reflects market sentiment: everyone is waiting for a new reshuffling of the rankings.

For marketing and content teams, this is especially important because AI tools for writing, image creation, research, and automation can change very quickly within just a few weeks. Instead of sticking to a single model, businesses should monitor workflow compatibility, cost, and real-world output quality.

Source: @LuminaXspace

The free path to becoming an AI engineer is still wide open

A resource roundup being shared by the community shows that anyone wanting to enter the AI engineer profession in 2026 can still learn almost the entire foundation from free sources: Stanford CS229, Harvard CS50 AI, MIT 6.S191, UC Berkeley LLM Agents MOOC, Hugging Face courses, plus content from Anthropic Academy and the Anthropic Prompt Engineering Course.

This is a very useful message for the Vietnamese market: the biggest barrier is no longer a “lack of materials,” but a lack of a consistent learning and practice roadmap. Marketing businesses can use these resources to upskill internal teams in AI prompting, automation, agent workflows, and LLM applications in daily work.

Source: @TheAIColony

Moonshot’s Kimi K3 makes a splash on the AI coding leaderboard

Moonshot AI is being mentioned because Kimi K3 has risen to the top of an important coding leaderboard in frontend coding, surpassing several big names such as Claude Fable 5 and GPT-5.6 Sol in the cited review. Alongside that is the story of a 2.8 trillion-parameter model and plans to release open weights for free in the near future.

However, the source also makes clear that this is still just one specific arena, with high uncertainty remaining and the main text leaderboard still showing the race is far from settled. Even so, what matters for enterprise users is that price and model openness are becoming major competitive advantages, especially for teams that need to self-host or customize deeply.

Source: @shanaka86

Grok 4.5 appears on Cognition’s new FrontierCode leaderboard

Cognition has just launched FrontierCode leaderboard, a ranking designed to measure which model can write code that can truly “merge into production.” According to the source description, Grok 4.5 has been added to this board alongside other leading models.

The new point here is that measurement is moving closer to what developers actually need: code must run, integrate cleanly, and solve real problems. For marketers using AI to create landing pages, automation scripts, or prototypes, the criterion of “usable in work” matters far more than a showy score.

Source: @teslaownersSV

The Ukraine personnel shake-up story is not the core AI angle of this roundup

Source number 7 mainly discusses a political development and questions the authenticity of a report related to Mykhailo Fedorov, without directly reflecting AI trends or products. Since the goal of this roundup is to compile AI updates with practical value for readers, we are not including this source as a main content item.

Editorial note: the source is acknowledged but not used in the analysis, to keep the article focused and avoid mixing in political news outside the topic scope.

Grok 4.5 is praised as very strong, and expectations are centered on the 2T model

Some social media comments describe Grok 4.5 as “impressive” and say the next big leap will come when the 2T model is released. In the accompanying quote, Elon Musk says the 2T model is in the final stage of initial training and could surpass Kimi while still keeping speed and token efficiency close to the current 1.5T model.

For the AI market, this is a familiar pattern: a large model announcement comes with very high expectations for quality, speed, and cost. What needs to be watched is not a single claim, but how stable the model is once it enters production, especially for teams that need long, multi-step, highly reliable workflows.

Source: @BrianRoemmele / @elonmusk

Anthropic brings Fable 5 into subscriptions: a sign that price pressure is rising

According to the cited announcement, Claude Fable 5 will be included in the Max and Team Premium plans, while Pro and Team Standard users will still access it through usage credits plus a one-time credit. Community reactions suggest this is a notable move because it gives more users access to a model that is highly rated for planning and handling complex tasks.

Behind this move is an important signal: the AI battle is not only about model capability, but also about monetization models. For businesses, this is the time to consider a multi-model strategy rather than relying entirely on a single provider for every need.

Source: @AlexFinn / @claudeai

Grok Build aims for an “AI working alongside you on your laptop” experience

Grok Build is described as an AI assistant that is fairly easy to install and can do many things on a personal computer: build apps, write and fix code, browse the web, research, automate workflows, operate production tools, and handle multi-step tasks from a single command.

The important point is that it does not require users to “speak like a programmer.” Users can describe their goal in natural language or by voice, then let the AI coordinate the remaining steps. For marketers, this is the kind of tool that can be very helpful for competitor research, campaign brief preparation, file organization, or automating repetitive daily tasks.

Source: @XFreeze

DeepSeek V4 is preparing for GA, with price seen as a major advantage

A closely watched update suggests DeepSeek V4 may soon move from preview to GA, with the v4 Pro version reaching 80.6% on SWE-bench, close to Opus 4.6’s 80.8% according to the cited figures. The source also highlights an output price of around 3.48 USD/million tokens versus 25 USD/million tokens for the named competitor, along with a 1M context window.

If these numbers hold at official release, it would be a reminder that the AI market is entering a very intense cost competition phase. For Vietnamese businesses, this is an opportunity to optimize AI budgets for large-scale tasks such as summarization, classification, customer support, or mass content variation generation.

Source: @Adidotdev

The AI race between the US and China is entering a “dual race”

The final comment in the news set argues that chip restrictions, export controls, and “distillation” accusations have not stopped Chinese models from advancing very quickly. The source highlights Zhipu and Moonshot as examples showing that open, cheaper, and customizable models are gradually building their own position, while closed US models still hold the advantage on many overall leaderboards.

What matters for product and marketing teams is that “winning” in AI is no longer a single-track race. One side may lead in overall quality, while the other wins on price, openness, and deployment flexibility. The near future will likely belong to teams that know how to choose the right model for the right job, rather than trying to stay loyal to just one ecosystem.

Source: @OopsGuess

A perspective for the Vietnamese market

For Vietnamese businesses, the stories above show three very clear trends. First, AI is moving from “chatbots” to execution agents that can plan, run tasks, and coordinate multiple steps. Second, cost and real-world deployability are becoming more important than model reputation. Third, the opportunity to learn and experiment with AI remains wide open thanks to a rich ecosystem of free resources and increasingly easy-to-use agent tools.

For marketing teams, a sensible priority over the next 6–12 months is to test AI for market research, draft content writing, report summarization, repetitive task automation, and building output review workflows. Those who know how to match the right model to the right workflow will gain a clear advantage in speed, cost, and productivity.

See more marketing news and guides at https://marketing365.vn.

Read more articles in the same category AI developments.

This article focuses on AI news with a perspective for the Vietnamese market.