The AI industry has never moved faster than it does right now.
What began as a surprising consumer hit with the launch of ChatGPT in late 2022 quickly escalated into a full-scale arms race, resetting global expectations for machine intelligence and forcing every major research lab to accelerate their roadmaps. Now, the competition between OpenAI and Anthropic has evolved into a high-stakes chess match played in public, where a breakthrough in one domain prompts an immediate counter-move in another.
OpenAI's latest move arrived with the release of GPT-5.5, a model positioned as a new class of intelligence built specifically for "real work" and autonomous agents.
Internally codenamed "Spud" and released on April 23, 2026, this model marks a pivot from simple dialogue toward practical execution.
It is engineered to grasp complex goals, use digital tools fluidly, and check its own outputs without constant human intervention. These improvements target high-intensity scenarios such as agentic coding across entire repositories, navigating operating systems, and conducting iterative scientific research.
GPT-5.5 excels at writing and debugging code, researching online, analyzing data, creating documents and spreadsheets, operating software, and moving across tools until a task is finished.
The gains are especially clear in agentic coding, computer use, knowledge work, and early…— OpenAI (@OpenAI) April 23, 2026
The technical shift in GPT-5.5 is centered on its ability to spawn autonomous sub-agents within unified environments like Codex.
This multi-agent orchestration architecture allows the model to move noisy, exploratory work off the main thread, solving the common problem of "context rot" in long conversations. Instead of processing a single linear prompt, the primary model acts as an orchestrator, spawning specialized workers such as an "explorer" for codebase mapping or a "reviewer" for security audits, which return distilled summaries to the main thread.
This agency is reflected in its performance on the Terminal-Bench 2.0 benchmark, where its 82.7% score outperformed Anthropic's Claude Opus 4.7 by a significant margin.
Efficiency remains a primary pillar of the 5.5 release, challenging the traditional scaling law where increased intelligence typically results in higher latency. GPT-5.5 maintains the same per-token latency as GPT-5.4 while proving to be substantially more token-efficient.
Although API pricing has shifted to $5 per million input tokens and $30 per million output tokens, the model often requires fewer total tokens to resolve complex reasoning problems.
This efficiency is further bolstered by its deployment on NVIDIA’s GB200 NVL72 infrastructure, which has enabled a 20% increase in token generation speeds compared to earlier configurations.
Further reading: With Computer Use On Mac, OpenAI Revives Codex As An 'Almost Everything' AI Agent
In ChatGPT, full-stack inference improvements enable a more capable model at faster speed. This efficiency is a game-changer for GPT-5.5 Pro, now a much more practical option for demanding tasks, and a step change in the level of difficulty and quality of work ChatGPT can take on…
— OpenAI (@OpenAI) April 23, 2026
This announcement landed just a week after Anthropic unveiled Claude Opus 4.7 on April 16, which sharpened performance in software engineering and precise instruction-following.
The competitive tension is particularly palpable in the realm of advanced mathematics; the GPT-5.5 Pro variant reached a 39.6% score on the FrontierMath Tier 4 benchmark, nearly doubling the performance of its closest rivals. This version introduces a specialized "Thinking" mode, allowing for increased internal compute time where the model verifies its own assumptions and traces complex logic before delivering a final response.
The rivalry extends into the sensitive domain of high-security models, where both labs maintain a guarded tier of intelligence.
Anthropic's Claude Mythos Preview remains limited to narrow partners for defensive work under Project Glasswing, cited for its potency in vulnerability discovery. OpenAI has pursued a parallel path with its GPT-5.4-Cyber variant, offering trusted access for legitimate defensive tasks. While public-facing models like GPT-5.5 and Opus 4.7 battle for dominance in everyday workflows, these restricted models define the true frontier of technical capability and safety.
GPT-5.5 is rolling out today for Plus, Pro, Business and Enterprise users across ChatGPT and Codex.
We’re also introducing GPT-5.5 Pro for Pro, Business, and Enterprise users in ChatGPT.— OpenAI (@OpenAI) April 23, 2026
Ultimately, GPT-5.5 feels like a deliberate escalation in a contest that shows no signs of slowing.
Each lab is responding not only to technical benchmarks but to shifting user expectations around autonomy and reliability.
Whether the next move comes from tighter enterprise integration or further refinements in agent persistence, the pattern is clear: progress arrives in rapid, public increments, and the machines are becoming increasingly capable of the complex work once reserved exclusively for humans.