# Gemini Is Back To The Frontier With Gemini 4 Argon > Source: > Published: 2026-10-01 08:09:24+00:00 A few weeks ago, it was fair to ask whether Google had lost the plot. The company that kicked off the modern AI era with the [transformer paper](https://officechai.com/stories/the-indian-researchers-whose-work-led-to-the-creation-of-chatgpt/) had slid down the leaderboards, its flagship model was missing deadlines, and its best public releases were all in the cheaper Flash tier. With Gemini 4 Argon, that story has changed. Artificial Analysis says Argon scores 53 on its Intelligence Index, level with OpenAI’s flagship and behind only Anthropic’s top models, which puts Google back in a tie for second among the labs. It’s a striking turnaround, and to appreciate it, you have to remember how bad the last few months looked. ## A Promise That Drew Groans At Google I/O on May 19, Google released Gemini 3.5 Flash and previewed its bigger sibling, Gemini 3.5 Pro. CEO Sundar Pichai told developers to give the company until the next month to get it to them, a line that reportedly drew an audible groan from the audience. Gemini 3.5 Flash was a solid agentic model, but it stopped short of frontier territory, and the Pro model that was meant to take Google back to the top was nowhere to be seen when June ended. The delays that followed were embarrassing. Bloomberg reported in mid-July, citing ten current and former employees, that Gemini 3.5 Pro was months behind schedule because of coding performance. Google had reportedly reset and refreshed its training data in late June to fix the problem, but the results were disappointing. Reports counted at least three missed internal deadlines, and the model sat in a limited enterprise preview. Eventually, it appeared that Gemini 3.5 Pro had been quietly shelved altogether, with Google choosing to skip ahead rather than ship a model leadership felt wouldn’t move the needle. ## The Flash Treadmill With no Pro model to offer, Google did what it could: it shipped Flash models, and kept shipping them. [Gemini 3.6 Flash and Gemini 3.5 Flash-Lite](https://officechai.com/ai/google-releases-gemini-flash-3-6-and-gemini-flash-3-5-lite/) arrived on July 21, a cheaper pair that came with no update on Pro’s timing. Gemini 3.7 Flash followed in mid-August, again with no word on the flagship. Then, on September 2, Google released [Gemini 3.8 Flash](https://officechai.com/ai/gemini-3-8-flash-scores-59-on-artificial-analysis-intelligence-index-jumps-3-points-over-gemini-3-7-flash/), which scored 59 on the Artificial Analysis Intelligence Index, a three-point gain over 3.7 Flash, at a cost of just $0.58 per task. On their own terms, these were respectable releases. Gemini 3.8 Flash was the fastest model on the chart and the cheapest at its level of intelligence. But the cadence told its own story. Flash [had become Google’s real release schedule](https://officechai.com/ai/google-is-internally-testing-out-gemini-3-8-flash-reports/), carrying a weight it was never designed to carry. Four Flash-tier releases in under four months, with no new Pro-class model, left Google with nothing to show against the top-end models from Anthropic and OpenAI. Its newest non-Flash model was still Gemini 3.1 Pro Preview, which dated back to February. ## Sliding Down The Leaderboard The rankings reflected it. Gemini 3.1 Pro Preview had briefly topped the Artificial Analysis Intelligence Index outright, holding the position for weeks before OpenAI and Anthropic caught up. But by the time Gemini 3.5 Flash landed, Google was [out of the top five AI labs on the index](https://officechai.com/ai/google-drops-out-of-top-5-ai-labs-on-artificial-analysis-intelligence-index-for-the-first-time/) for the first time, behind Meta’s Muse Spark and SpaceXAI’s Grok among others. From there, the slide continued, and Google [kept dropping down the table](https://officechai.com/ai/google-slips-to-10th-place-among-ai-labs-on-the-artificial-analysis-intelligence-index/) as rivals shipped and index updates re-anchored the scores. At its lowest, Google sat in ninth place, behind a long line of competitors including labs that few would have put ahead of it a year ago. The criticism got louder, too. Rivals openly mocked Gemini’s position on the index, and Google was left defending its Flash-first strategy while the Pro-sized hole in its lineup grew more obvious. ## Argon Changes The Picture Gemini 4 Argon is the first Google proprietary model above the Flash class in more than seven months, and its debut flips the narrative. At high reasoning, Argon scores 53 on the latest version of the index (v4.3.2), matching [GPT-6 Astra](https://officechai.com/ai/gpt-6-astra-benchmarks/) at 53 and sitting one point ahead of GPT-6.1 Sol at 52. Only Anthropic is clearly ahead: Claude Opus 5.5 scores 58 and Claude Sonnet 5.5 scores 56. That leaves Google tied for second among the labs, jumping from ninth to the top tier in a single release. The size of the jump is the real story. Argon is 23 points above Gemini 3.1 Pro Preview, which scored 30 on the same index, and 12 points above Gemini 3.8 Flash. (Note that the index has been revised several times this year, which is why the Flash scores of 56 to 59 seen on earlier versions aren’t directly comparable to Argon’s 53.) Artificial Analysis credits lower hallucination rates and better agentic performance. Argon’s 15% hallucination rate on AA-Omniscience is the lowest of any model scoring 45 or more, and it takes the top spot on AutomationBench-AA at 77.5%, a benchmark where Google’s models have historically lagged. ## Not A Victory Lap Yet There are caveats worth keeping in mind. Argon isn’t publicly available yet; it’s rolling out to selected users, with a broader launch to come. Its attractive pricing is a temporary 50% promotion off a standard $4/$20 per million tokens, and Google hasn’t said when the discount ends. Anthropic still holds the top of the chart, and on terminal-based coding tasks, Argon continues to trail several rivals. But for a company that spent the summer missing deadlines and falling down the rankings, the direction of travel is important. Google went from shipping stopgaps to matching OpenAI’s flagship on an independent index, and did it while its own Pro line was in disarray. The race at the top is now a three-way contest again, and Gemini is back in it.