Gemini faces trust test as users report weaker reasoning and coding

Google’s claims of rapid Gemini progress are being challenged by users reporting inconsistent reasoning, poorer coding and a widening gap between benchmark results and everyday performance.

Google’s Gemini chatbot is facing renewed scrutiny as complaints about weaker reasoning and coding performance continue to mount. What began as anecdotal frustration from users on forums and social platforms has grown into a wider debate about whether the model has become less reliable in day-to-day use, even as Google continues to present it as a fast-improving product. Android Authority has highlighted the issue in a detailed discussion that captured many of the concerns now circulating among developers and regular users.

The criticism centres on a familiar tension in modern AI development: models can look stronger in carefully selected benchmarks while feeling less capable in ordinary tasks. Users say Gemini’s earlier versions were better at following multi-step logic, maintaining context across longer exchanges and producing cleaner code. More recent iterations, by contrast, are being described as prone to contradictions, shallow explanations and occasional refusals to answer straightforward prompts. That mismatch between official progress claims and lived experience is at the heart of the backlash.

Coding performance has become one of the most sensitive fault lines. The Los Angeles Times reported in July that Google’s internal work on Gemini 3.5 Pro had been slowed by difficulties improving programming ability, with engineers reportedly frustrated by the pace of progress. Separately, Windows Central described a recent Gemini incident in which the chatbot became trapped in an error loop while trying to code a compiler, even calling itself a “fool” and a “disgrace”. Google said it would fix the problem, but the episode reinforced the sense that the system still struggles with complex software tasks.

That concern matters because coding is now one of the clearest commercial uses for large language models. TechRadar has argued that progress in code generation has not been matched by comparable gains in software security, with AI-produced code still capable of introducing vulnerabilities and unsafe assumptions. In that context, even small regressions are significant. For developers who rely on Gemini to draft scripts, explain functions or troubleshoot errors, a model that is less consistent than before can quickly become a liability rather than a productivity tool.

Google is also dealing with pressure on another front: protecting the model itself. Computerworld reported that the company blocked a coordinated attempt involving more than 100,000 prompts designed to extract proprietary reasoning traces from Gemini and potentially clone its behaviour. That episode shows how valuable the model’s internal reasoning has become, and how aggressively rivals and researchers are probing the systems behind it. It also underlines the difficulty of balancing openness, performance and security in a fiercely competitive market.

The competitive backdrop is becoming more unforgiving. Tom’s Guide recently reported that a small start-up, Poetiq, outperformed Gemini 3 on a reasoning benchmark by using a meta-system that refines outputs from existing models. The result suggested that smaller or more targeted approaches can sometimes rival the scale-heavy strategies adopted by major technology companies. For Google, that is a warning sign. If users already believe Gemini is slipping in practical quality, and competitors are still making visible gains, confidence can erode quickly.

Google continues to insist that Gemini is improving overall, but the company now faces a more difficult test than a single benchmark score. It must prove that the model can remain dependable across real-world tasks, not merely competitive in laboratory settings. For users, the issue is less about headline numbers than about whether the chatbot can reason clearly, code accurately and stay coherent over time. Until that gap is narrowed, the criticism is likely to persist.