May 2026 Tech Roundup: GPT-5.5 Arrives, Waymo’s Ojai Goes Public, and Biotech’s New Benchmarks

In the last month, the pace of progress across AI, autonomous vehicles, and biotechnology accelerated sharply. OpenAI shipped GPT-5.5, Google pushed Gemini 3.5 deeper into agentic workflows, and Waymo began carrying passengers in its next-generation robotaxi. In biotech, a new DNA-synthesis technique and a landmark pancreatic-cancer drug trial redefined what rapid iteration can mean for medicine. Here is the month’s most consequential news, with context on why each development matters.

The AI Model Rush Just Got Real

1. OpenAI ships GPT‑5.5

OpenAI’s April release of GPT‑5.5 marks another inflection point in the race toward agentic AI. The model is positioned not as another incremental upgrade, but as a step change in getting work done on a computer. OpenAI says GPT‑5.5 is its smartest and most intuitive model, with especially strong gains in agentic coding, computer use, knowledge work, and early scientific research.

The benchmark numbers back that up. On Terminal‑Bench 2.0, GPT‑5.5 posts 82.7 %, compared with 75.1 % for GPT‑5.4. OSWorld‑Verified climbs from 75.0 % to 78.7 %, BrowseComp from 82.7 % to 84.4 %, and FrontierMath Tier 1–3 from 47.6 % to 51.7 %. A GPT‑5.5 Pro variant pushes some scores even higher and is now available to Pro, Business, and Enterprise users in ChatGPT, with API access rolling out afterward.

Equally important is efficiency. OpenAI says GPT‑5.5 matches GPT‑5.4 per‑token latency while using significantly fewer tokens to complete the same Codex tasks—a win for both cost and speed. The release also comes with a refreshed safety stack: red‑team testing, advanced cybersecurity and biology evaluations, and nearly two hundred trusted early‑access partners providing feedback before launch.

2. Claude Opus 4.8 and Google Gemini 3.5 Flash

While OpenAI occupied headlines, Anthropic quietly upgraded Claude Opus to 4.8 on 28 May, refining the performance and usability gains of its predecessor. On the Google side, Gemini 3.5 Flash is being marketed as frontier performance for agents and coding—specifically targeting long‑horizon, real‑world utility tasks. Google claims 3.5 Flash outperforms Gemini 3.1 Pro on Terminal‑Bench 2.1 (76.2 %), GDPval‑AA (1656 Elo), and MCP Atlas (83.6 %), while running at four times the output‑token speed of other frontier models.

What makes the timing notable is the convergence around the same use case: software engineering, multi‑step workflows, and agentic behavior. Gemini is being deployed through Google’s Antigravity harness and developer platform; OpenAI is pushing Codex deeper into agency; Anthropic continues to refine Opus for complex reasoning. Competition is healthy, but the category is also sharpening—the ground is shifting from chat to acting.

3. MiniMax M3, Mistral Medium 3.5, and the open‑model ecosystem

Other important releases in this cycle include MiniMax M3, a frontier coding model with a one‑million‑token context window and native multimodality, and Mistral Medium 3.5, which powers remote coding agents in Vibe. Tencent also released Hy3, an open Mixture‑of‑Experts model targeting agent capabilities and real‑world usability. NVIDIA shipped its Alpamayo 2 Super open reasoning model alongside a new Agent Toolkit for enterprise partners. Taken together, the open‑model landscape has become surprisingly competitive with the closed frontier labs, offering cheaper or locally deployable alternatives.

Autonomous Driving Enters the Build Phase

Waymo’s Ojai robotaxi begins passenger rides

Late in May, Waymo started inviting select passengers into its newest robotaxi: the Zeekr RT minivan, rebranded as Ojai. Riders in San Francisco, Los Angeles, and Phoenix now have access, with trips offered free initially while Waymo collects passenger‑experience data before introducing paid fares.

The Ojai is more than a refresh. It is the debut of Waymo’s sixth‑generation autonomous system, succeeding the fifth‑gen platform that has run on Jaguar I‑Paces since 2020. Waymo says the new system is smarter and more capable, yet uses fewer sensors to lower manufacturing costs. The cameras, lidar, and radar are improved individually, but the larger story is production scale: Waymo claims its manufacturing partners can produce tens of thousands of units a year.

The partnership with Chinese automaker Geely (Zeekr’s parent) is itself a signal. Waymo says it strips the vehicles of connected software before importing them, sidestepping U.S. restrictions on Chinese vehicle software. The result is a Chinese‑built, software‑neutral chassis purpose‑designed for autonomy, which says something about how the global supply chain for robotaxis is evolving.

World models and the autonomy layer

Underlying the vehicle rollout is an equally important software layer. Xiaomi EV introduced its world model to advance autonomous driving tech; NIO announced a June OTA upgrading its proprietary NIO World Model to more human‑like smart driving. These world models are not purely perception—they simulate future states, predict other agents’ behavior, and plan accordingly. In effect, the industry is moving from reactive driving systems to cognitive ones.

Tesla is pushing forward as well, with reports in early June confirming FSD version 14.3.3 as part of the Spring 2026 update, alongside new infotainment features like Hey Grok and Pet Mode. And Autobrains, working with Uber, plans an agentic AI robotaxi program in Munich built on NVIDIA DRIVE Hyperion. The autonomous‑driving market is consolidating around a few core themes: cheaper sensor suites, simulation‑trained reasoning, and high‑volume manufacturing.

Biotech: From AI‑Generated Genomes to Cancer Pills

Sidewinder rewrites the rules of DNA synthesis

One of the most under‑reported stories this month was the introduction of Sidewinder, a DNA‑assembly technique that could speed up the transition from AI‑designed genomes to real, lab‑built organisms. The method can assemble dozens of genetic sequences simultaneously in a single test tube, with error rates of roughly one incorrect junction per ten million assembly events—an order of magnitude more accurate than conventional methods, which misfire roughly once every ten to thirty joins.

The implications are concrete. Researchers led by Caltech synthetic biologist Kaihang Wang used Sidewinder to build, from scratch and with zero errors, a 12,500‑letter DNA sequence of the E. coli genome that had been redesigned in silico by the generative model Evo 2. Sequences of that length can encode entire biochemical pathways, laying the groundwork for engineered microbes that manufacture drugs, biofuels, or specialty chemicals.

Bioengineer Thomas Gorochowski at the University of Bristol called it a step change, noting that synthesizing large genetic systems, and eventually small genomes, is central to the generative‑biology revolution that AI is enabling. The preprint is already circulating, and Sidewinder’s reliance on cheap raw materials could democratize access to large‑scale DNA construction outside deep‑funded pharma labs.

Daraxonrasib and the KRAS breakthrough

In oncology, a daily pill called daraxonrasib nearly doubled median survival for advanced pancreatic cancer patients in a phase 3 trial led by American scientists and presented at the American Society of Clinical Oncology meeting in Chicago. The trial involved 248 patients on daraxonrasib and 252 on chemotherapy, most with KRAS‑mutated tumors. Median survival time in the drug arm was substantially longer, and severe side effects occurred in 43.6 % of patients versus 57.5 % on chemotherapy.

Pancreatic cancer remains one of the deadliest cancers: more than half of patients die within three months of diagnosis in the UK, and just over eleven thousand cases are recorded annually there. Experts, including Rachna Shroff of the University of Arizona Cancer Center and Anna Jewell of Pancreatic Cancer UK, described the data as landscape‑changing. If the drug clears regulatory review and enters routine care, it would represent one of the most meaningful oncology advances in decades.

CRISPR safety, mRNA vaccines, and gene therapy

Elsewhere in biotech, a team at the San Raffaele Telethon Institute for Gene Therapy published a new SMArT platform designed to improve the safety of CRISPR gene editing. Separately, GSK announced that bepirovirsen achieved unprecedented functional cure rates in chronic hepatitis B, with the potential to redefine treatment standards. Voyager Therapeutics received FDA IND clearance for VY1706, a gene therapy aimed at reducing tau production in the brain for Alzheimer’s disease—a novel mechanism targeting protein accumulation rather than amyloid alone.

And in yet another sign of mRNA’s versatility beyond Covid‑19, researchers reported progress in personalized mRNA cancer vaccines targeting melanoma, with elderly volunteers participating in early trials. Microbiologist commentary has framed these advances as a maturation of the platform—faster prototyping, cleaner editing, and more predictable outcomes.

Why This Matters

These three domains—frontier AI, autonomous transport, and biotech—are not running on separate tracks. Sidewinder’s ability to rapidly build AI‑generated DNA blueprints creates a direct feedback loop between generative models and synthetic biology. GPT‑5.5’s agentic capabilities accelerate software development, the codebase that powers the simulation environments used to train self‑driving world models. And the same underlying advances in chip density and sensor technology make cheaper manufacturing of both robotaxi lidar and high‑throughput lab equipment possible.

The common thread is acceleration: the time between an idea, a model, a blueprint, or a molecular structure and its physical realization is compressing. For engineers, founders, and researchers, staying current means tracking these clusters together, not in isolation.