AI Business and Development Daily News Rundown March 05th 2026: GPT-5.4 Drops: Native Computer Use and the 1M Token Breakthrough

Full Audio at

This episode is made possible by our sponsors:

🎙️ Djamgamind: Information is moving at the speed of light. Djamgamind is the platform that turns complex mandates, tech whitepapers, and clinic newsletters into 60-second audio intelligence. Stay informed without the eye strain. 👉 Get Your Audio Intelligence at DjamgaMind.com

Summary: OpenAI accelerates its release cycle with the launch of GPT-5.4, a model built for “Agentic Workflows.” Featuring native computer use, an 83% score on professional knowledge benchmarks (GDPval), and a massive 1-million token context window, this update marks a major shift toward automated professional services.

Key Points:

• Native Computer Use: GPT-5.4 can now operate desktop software, websites, and apps via mouse and keyboard commands.

• Professional Benchmark (GDPval): Achieved an 83% success rate across 44 professional occupations, outperforming human experts in many fields.

• Accuracy Leap: 33% reduction in factual hallucinations compared to GPT-5.2.

• Two New Tiers: GPT-5.4 Thinking (for ChatGPT reasoning) and GPT-5.4 Pro (for high-throughput API/Enterprise tasks).

• 1M Context Window: Now available in Codex and the API for massive dataset analysis.

Keywords: GPT-5.4, OpenAI Thinking, Computer Use AI, GDPval Benchmark, AI Agents, 1M Context Window, Professional AI, Tech News 2026.

🚀 Reach the Architects of the AI Revolution

Want to reach 60,000+ Enterprise Architects and C-Suite leaders? Download our 2026 Media Kit and see how we simulate your product for the technical buyer: https://djamgamind.com/ai

Connect with the host Etienne Noumen: https://www.linkedin.com/in/enoumen/

🎙️ Djamgamind: Information is moving at the speed of light. Djamgamind is the platform that turns complex mandates, tech whitepapers, and clinic newsletters into 60-second audio intelligence. Stay informed without the eye strain. 👉 Get Your Audio Intelligence at

https://djamgamind.com/

⚗️ PRODUCTION NOTE: We Practice What We Preach.

AI Unraveled is produced using a hybrid “Human-in-the-Loop” workflow. While all research, interviews, and strategic insights are curated by Etienne Noumen, we leverage advanced AI voice synthesis for our daily narration to ensure speed, consistency, and scale.

GPT-5.4 Drops: Native Computer Use and the 1M Token Breakthrough

OpenAI (yes, them again!) rolled out two brand new models on Thursday, GPT 5.4 Thinking and GPT 5.4 Pro. The first is a reasoning model aimed at complex work, like coding and supervising AI agents. OpenAI specifically noted that GPT 5.4 Thinking is optimized for running agents with less computing power, meaning lower costs for Claw-pilled users like Jason. Pro, on the other hand, is designed strictly for “maximum performance.”

OpenAI also notes that GPT 5.4 is its most “factual” model to date: 18% less likely to make errors, and 33% less likely to hallucinate, than GPT 5.2. (You should probably still ask Claude or Grok for double-confirmation.) It’s also more adept at handling responses that require citing multiple sources. GPT 5.4 is headed to Codex and the OpenAI API, while the “Thinking” version is en route to ChatGPT.

For those keeping track, it’s only been a few days since OpenAI brought us GPT 5.3 Instant, a smoother and faster model designed for “everyday conversations.” It’s neat to have so many new models dropping so frequently, but AI companies should be on guard against fatigue, lest a new model launch become a non-event.

Anthropic reopens Pentagon talks over Claude AI deal

• Anthropic has reopened negotiations with the Pentagon over a deal that would define how the U.S. military can access and deploy its Claude AI models after talks collapsed last week.

• The breakdown centered on a dispute over bulk data analysis language, with Anthropic pushing for guarantees against domestic surveillance and autonomous weapons while the Pentagon wanted fewer restrictions.

• OpenAI announced a new Defense Department deal hours after the White House criticized Anthropic, though CEO Sam Altman later said his company “shouldn’t have rushed” that agreement.

Amodei memo rips OAI-Pentagon deal as ‘safety theater’

Anthropic CEO Dario Amodei sent an internal memo on Friday calling OpenAI’s Pentagon AI deal “maybe 20% real and 80% safety theater” while also taking direct shots at Sam Altman in the 1,600-word document released by The Information.

The details:

• The Pentagon labeled Anthropic a “supply chain risk” last week, followed immediately by OAI agreeing to its own Dept of War deal with similar terms.

• Amodei accused Altman of “gaslighting” and pointed to Greg Brockman’s $25M Trump donation while Anthropic refused to give “dictator-style praise”.

• He also said OAI is “trying to spin this as we were unreasonable… didn’t engage in a good way, less flexible”, calling it a pattern he’s seen often from Altman.

• Amodei also struck a different tone on Tuesday with regard to the Pentagon, saying the two sides “have much more in common than we have differences.”

Why it matters: Dario’s memo comes in HOT and reads like someone letting loose after holding back since he left OpenAI in 2020. Between this, the hand-hold refusal, and Super Bowl ads before that, the frontier rivalry has never been more personal — but the Pentagon deal is also turning into a big mess for all parties involved.

OpenAI building its own GitHub to ditch Microsoft

OpenAI is reportedly building an internal code repository platform to replace Microsoft’s GitHub, according to The Information — a move that could pit it (once again) against one of its biggest financial backers.

The details:

• The project kicked off after GitHub suffered repeated outages tied to an ongoing migration of its infrastructure, frustrating OpenAI engineers.

• GitHub’s CTO told staff the full Azure move would take two years, and several dev teams got reassigned to migration duty.

• OAI staffers have floated opening the platform to outside paying customers, pairing with Codex coding agents to consolidate the entire process.

Why it matters: This relationship continues to be a strange one, with OAI continuing to encroach on territory its largest investor already owns — from Office tools to the platform where 100M+ devs store their code. Microsoft is in the awkward position of funding a partner that is hammering away at the legacy tech giant’s developer moat.

Big Tech signs White House pledge to cover AI energy costs

• Seven major tech companies — Google, Meta, Microsoft, Amazon, Oracle, xAI, and OpenAI — signed a voluntary White House pledge to cover the energy costs their data centers impose on regular electricity customers.

• The pledge carries no penalties and the White House has no jurisdiction over state utility commissions that actually set power rates, meaning enforcement falls to the same local regulators already struggling with rising bills.

• Anthropic, which was excluded from the ceremony after being designated a “supply chain risk,” had already made the most concrete commitment of any company, pledging to cover 100% of consumer price increases caused by its data centers.

China five-year plan prioritizes AI and chip breakthroughs

• China’s latest five-year plan mentions AI more than 50 times and lays out goals to adopt the technology across its economy while pushing for breakthroughs in chips, quantum computing, and humanoid robots.

• The plan includes an “AI+ action plan” with specific measures like deploying robots in sectors facing labor shortages and using AI agents that can perform tasks with minimal human guidance.

• For the first time, the blueprint highlights open-source AI as a flagship strategy, which analysts say marks a key difference between China’s and America’s competing approaches to the technology.

Apple Music will tag up AI-generated tracks

• Apple Music is introducing new metadata tags that let record labels and distributors flag when AI-generated or AI-assisted content is part of a song uploaded to the platform.

• The tags let distributors mark specific parts of a release — including artwork, track, composition, or music video — to show where AI was involved in the creation process.

• The system is opt-in, meaning labels and distributors must manually choose to flag their use of AI, which is a similar approach to what Spotify is doing.

Apple finds new way to spot AI hallucinations

Apple may not have homegrown AI. But it wants to make sure the technology is done right.

On Tuesday, Apple published research detailing a new way to find and quash incidents of hallucination, the pesky mistakes that an AI model makes when it doesn’t have enough training data and starts making guesses. Apple’s research introduces “Reinforcement Learning for Hallucination Span Detection,” which pinpoints not just when an AI model hallucinates, but where exactly within a line of text the model goes wrong.

Apple’s model gives its AI framework small rewards each time it accurately identifies incorrect phrases or words, based on how closely its responses match those of human evaluators.

• This turns hallucination detection from a “binary task” into a “multi-step decision-making process,” Apple said in its research.

• To put it simply, it’s the difference between a teacher saying you failed a test with no explanation and a teacher telling you exactly which answers you got wrong and why.

“Most existing research works focus on a binary hallucination detection problem, where the goal is to determine if the model output contains hallucinations or not,” Apple said in the paper. “While useful, this formulation is limited: in many real-world applications, one often needs to know which specific spans in the model output are hallucinated in order to assess the reliability of the generated content.”

And Apple’s system proved itself, outperforming conventional methods on the RAGTruth Benchmark, an AI truth-checking test for tasks like summarization, question answering, and data-to-text.

New framework tests if AI helps students learn

Image source: OpenAI

OpenAI published a new framework built with Stanford and Estonia’s University of Tartu to track whether ChatGPT helps people learn over time — the company’s first attempt to measure AI’s impact on how students retain knowledge.

The details:

• OAI’s Learning Outcomes Measurement Suite is designed to track AI’s impact on learning over time, measuring factors like motivation, persistence, and recall.

• A 300+ student trial found microeconomics students scored 15% higher with ChatGPT’s study mode, though other subjects weren’t statistically significant.

• Estonia is running the framework’s biggest test, with 20K high schoolers across the country being tracked for a full semester.

Why it matters: AI has already made a serious impact in learning on both ends — from positive results like the Alpha School’s AI-first programming to negative ones like studies showing AI’s ‘brainrot’ impact on critical thinking. But no matter where the data ends up, it’s going to be a major part of the future of education either way.

What Else Happened in AI on March 05th 2026?

Nvidia CEO Jensen Huang said the chipmaker’s $30B OAI investment will likely be its last before the AI giant goes public, calling the original $100B deal “not in the cards”.

OpenAI launched the Codex app on Windows, featuring a native sandbox to let AI agents work directly in Windows-specific environments.

The White House announced a “Ratepayer Protection Pledge” signed by major AI companies, committing data center operators to fund their power and grid upgrades.

Google is facing a wrongful-death lawsuit from a parent alleging its Gemini chatbot developed an emotional relationship with his son and encouraged self-harm.

Elon Musk posted on X that “Tesla will be one of the companies to make AGI and probably the first to make it in humanoid/atom shaping form”.