OpenAI released GPT-5.5 on 23 April 2026, delivering 40% fewer tokens for equivalent coding tasks while targeting agentic workflows in science, research, and drug discovery.
Three-quarters of AI's economic gains are captured by just 20 percent of companies, according to PwC's 2026 AI Performance Study — and OpenAI's release of GPT-5.5 on 23 April suggests the gap is about to widen further, as the model introduces efficiency and capability advantages that will take most organizations months to deploy meaningfully.
GPT-5.5 arrived with a headline metric that matters to developers and enterprises more than benchmark scores: it completes equivalent Codex tasks using 40 percent fewer tokens than its predecessor, GPT-5.4. That efficiency gain translates directly into operating cost. For companies running thousands of automated code reviews, test generation runs, or document analysis pipelines daily, 40 percent fewer tokens means roughly 40 percent lower inference bills at the same output quality — or the same budget buying substantially more throughput.
OpenAI described GPT-5.5 as designed for "agentic" tasks — workflows in which the model must take sequences of actions, use tools, and make decisions without constant human input. The company cited agentic coding, computer use, scientific research, and drug discovery workflows as target use cases. The model is available to ChatGPT Plus, Pro, Business, and Enterprise subscribers, and the API versions — GPT-5.5 and GPT-5.5 Pro — became available to developers on 24 April. OpenAI said the model was priced higher than GPT-5.4 on a per-unit basis but delivered better output quality per token spent.
“OpenAI described GPT-5.5 as designed for "agentic" tasks — workflows in which the model must take sequences of actions, use tools, and make decisions without constant human input.”
The release was notable for what accompanied it: extensive third-party red-teaming for cyber and biosecurity risks, conducted before launch. OpenAI disclosed this process publicly — a response, in part, to growing regulatory pressure from the European Union's AI Act and the Trump administration's executive orders on AI procurement, which require some level of safety evaluation disclosure for frontier models.
Key Takeaways
→openai gpt-5.5: GPT-5.
→artificial intelligence 2026: GPT-5.
→chatgpt update: GPT-5.
→agentic ai: GPT-5.
The same 24 hours also produced a rival launch. DeepSeek, the Chinese AI lab that rattled markets with its V3 release in January 2026, unveiled V4 Flash and V4 Pro on 24 April, claiming top-tier open-source performance and introducing what it called a "Hybrid Attention Architecture" for improved long-context memory. DeepSeek's releases have consistently arrived at lower operating costs than OpenAI's equivalents, and V4 Pro's performance claims — if substantiated by independent benchmarks — would put it within striking distance of GPT-5.5 at a fraction of the API cost. OpenAI has publicly acknowledged DeepSeek as the most credible competitive threat outside the Google ecosystem.
The broader competitive backdrop also shifted this week. Google Cloud unveiled its latest-generation TPU chips between 20 and 22 April, positioning them as a challenge to Nvidia's dominance of AI training and inference hardware. Meanwhile, Anthropic holds the top position in independent AI model performance rankings as of March 2026, according to rankings published by research organizations tracking frontier model capability — ahead of xAI, Google, and OpenAI. That ordering has shuffled repeatedly over the past 18 months as each lab releases new versions on an increasingly compressed schedule. GPT-5.5 is OpenAI's second major release within a matter of weeks, reflecting a cadence the company has not publicly sustained before.
Advertisement
What GPT-5.5 means for businesses depends almost entirely on whether they are among that 20 percent of organizations PwC found capturing most of AI's gains. The efficiency improvement helps companies already operating at scale — those with established data pipelines, AI engineering teams, and deployment infrastructure. For the 80 percent that have so far struggled to move beyond pilots and proof-of-concept tools, a more capable model does not solve the organizational problem. Satya Nadella, Chief Executive of Microsoft, said at a conference in San Francisco on 14 April 2026 that "the bottleneck is not the models; it's the ability of enterprises to actually change how they work." That observation applies directly to GPT-5.5: the technology is ready; the question is whether the organizations using it are.
The next significant date for OpenAI watchers is the company's developer conference, scheduled for late May 2026, where the company is expected to outline its roadmap for GPT-6, codenamed "Spud," which entered pretraining earlier this year. Whether GPT-5.5 proves to be a competitive hold against DeepSeek V4 or merely the latest move in a race without a finish line will become clearer as independent benchmarks publish over the coming weeks.
GPT-5.5 is OpenAI's latest AI model, released on 23 April 2026. It is available to ChatGPT Plus, Pro, Business, and Enterprise subscribers, with API access (GPT-5.5 and GPT-5.5 Pro) opening on 24 April 2026.
How much more efficient is GPT-5.5 than its predecessor?
OpenAI says GPT-5.5 completes equivalent Codex coding tasks using 40% fewer tokens than GPT-5.4. For organizations running large-scale AI pipelines, this translates to roughly 40% lower inference costs for the same output quality.
How does GPT-5.5 compare to DeepSeek V4?
DeepSeek released V4 Flash and V4 Pro on 24 April 2026, the same week as GPT-5.5. DeepSeek's models have historically offered lower operating costs than OpenAI's equivalents. Independent benchmarks comparing V4 Pro to GPT-5.5 were not yet published as of 25 April 2026.
Who leads AI model performance rankings in 2026?
As of March 2026, Anthropic leads independent AI model performance rankings, followed by xAI, Google, and OpenAI, according to research organizations tracking frontier model capability. The rankings have shifted repeatedly as each lab releases new versions on a compressed schedule.
What safety testing did OpenAI conduct before releasing GPT-5.5?
OpenAI disclosed that GPT-5.5 underwent extensive third-party red-teaming for cybersecurity and biosecurity risks before its release on 23 April 2026 — a transparency measure aligned with requirements under the EU AI Act and Trump administration executive orders on AI procurement.
