OpenAI Defaults Free Users to ‘Instant’ Model to Cut Cost

OpenAI has quietly altered how millions of free users access ChatGPT. To stem a projected $14 billion deficit, the company is now defaulting non-paying accounts to its cheapest “Instant” model, removing the automatic routing that previously granted access to higher-reasoning capabilities.

While release notes frame the shift as “maximizing choice,” the move effectively restricts free access to the more expensive “Thinking” models. Users must now manually toggle settings to access advanced reasoning, a friction point designed to reduce compute costs as the company battles Google’s Gemini 3 Pro.

The ‘Instant’ Default: Economics Disguised as UX

Far from a simple UI tweak, the removal of automatic routing represents a fundamental shift in how OpenAI manages its massive free-tier user base. Under a recent update discovered by Wired, accounts on the Free and “Go” plans are now served by the GPT-5.2 “Instant” model by default:

Previously, the system would dynamically assess a user’s query and route complex tasks to a more capable “Thinking” model if necessary. According to the official documentation, that automated assistance has been dismantled.

“Previously, some questions were automatically routed to the Thinking model when ChatGPT determined it might help. To maximize choice, free users will now use GPT-5.2 Instant by default, and can still choose to use reasoning anytime by selecting Thinking from the tools menu in the message composer.”

By forcing users to manually select the “Thinking” option from a tools menu for every interaction requiring depth, OpenAI introduces a deliberate friction point. This UI change aligns directly with the company’s urgent need to control inference costs.

Running the “Instant” model costs approximately $1.75 per million input tokens, a fraction of the compute resources required for reasoning-heavy models. For a user base numbering in the hundreds of millions, steering default traffic to this lower-cost tier could save millions in daily operational expenses.

Framing this downgrade as a user-centric feature, the release notes suggest the change is intended to “maximize choice.” However, the practical result is that casual users will interact almost exclusively with the lower-fidelity model unless they actively intervene.

Frequent users have previously criticized the “lobotomized” feel of lower-tier models, a sentiment the company has struggled to address while managing server load. CEO Sam Altman recently validated these user complaints with a blunt admission on social media.

Code Red: The Strategic Pivot

Driving this shift is a financial reality that has forced the company into a wartime operational footing. Following the rapid ascent of Google’s Gemini 3 Pro to 650 million Monthly Active Users (MAUs), Altman declared a company-wide ‘Code Red’ emergency.

Facing a rival that was iterating faster and capturing market share with its multimodal capabilities, leadership has been forced to make hard trade-offs between innovation and survival.

Internal projections paint a concerning financial picture, estimating a $14 billion net loss for 2026 if current spending trajectories continue. To bridge this gap, the company is aggressively reallocating engineering talent and compute resources away from experimental features and toward core model stability.

Casualties of this prioritization include the highly anticipated “Pulse” personal assistant, which has been delayed indefinitely. Similarly, ChatGPT ‘Adult Mode’ features, originally slated for a December release to expand creative utility, have been pushed to Q1 2026.

Fidji Simo, CEO of Applications, clarified that the emergency status was intended to align the workforce rather than induce panic.

“We announced this code red to really signal to the company that we want to martial resources in one particular area, and that’s a way to really define priorities and define things that can be deprioritized.”

To stabilize the ship, the company has also reshuffled its executive deck. Signaling a move toward rigorous enterprise monetization, the appointment of Denise Dresser as Chief Revenue Officer brings a “SaaS-ification” mindset to the organization. The former Slack CEO is tasked with building the sales motion required to offset the significant burn rate.

The Price War: Undercutting Claude and Gemini

Beyond internal cost-cutting, OpenAI is using the GPT-5.2 release to wage an aggressive price war against its primary rivals. The new model series, which includes “Instant,” “Thinking,” and “Pro” variants, features a unified API pricing structure designed to undercut the premium tier of the market.

Priced at $1.75 per million input tokens, GPT-5.2 significantly undercuts the Claude Opus 4.5 launch, which Anthropic priced at roughly $5.00 per million input tokens. It also competes directly with Google’s Gemini 3 Pro, which hovers around the $2.00 mark for standard contexts.

On the output side, the disparity is even more pronounced. GPT-5.2 charges $14.00 per million output tokens, compared to $25.00 for Claude Opus 4.5. This pricing strategy aims to neutralize the cost advantage that Google has historically leveraged through its TPU infrastructure.

Technical validation for the new series comes from the “Thinking” model’s performance on the industry-standard SWE-bench Verified. Achieving a score of 80.0%, the model effectively matches the 80.9% score of Claude Opus 4.5 and surpasses the 76.2% score of Gemini 3 Pro.

To further differentiate its offering, OpenAI has introduced “GDPval,” a proprietary benchmark designed to measure economic utility rather than abstract reasoning.

“We designed GPT‑5.2 to unlock even more economic value for people; it’s better at creating spreadsheets, building presentations, writing code, perceiving images, understanding long contexts, using tools, and handling complex, multi-step projects.”

According to this internal metric, the model achieves a 70.9% win/tie rate against human experts across 44 distinct occupations.

However, independent analysts remain skeptical of such vendor-created benchmarks, noting that they lack the transparency and reproducibility of open standards like SWE-bench. While OpenAI executives have touted these metrics, independent third-party verification of the GDPval results remains unavailable.

Personalization as a Retention Tool

To mitigate the friction of the “Instant” model downgrade, OpenAI is simultaneously rolling out granular personalization features.

According to the documentation, the automated assistance that previously smoothed the user experience has been dismantled. Where the system once intelligently detected complex queries and seamlessly routed them to the more capable “Thinking” model, it now defaults strictly to the entry-level GPT-5.2 “Instant” tier.

While OpenAI frames this shift as a move to “maximize choice,” the practical outcome is a new layer of friction: users must now navigate to the tools menu and manually select the reasoning capabilities for every specific interaction that requires them, rather than relying on the AI to adapt on the fly.

These controls appear designed to deepen user engagement and retention, offering a “stickier” experience even as the underlying model for free users becomes less capable.

By allowing users to tailor the “vibe” of the interaction, the company hopes to maintain the conversational intimacy that differentiates ChatGPT from Google’s more utilitarian interface.

Trust remains a fragile commodity, however. The rollout follows a backlash about ads-like retailer links in chats, where criticism forced the company to disable “app suggestions” that were perceived as hidden advertisements.

Moving forward, the monetization strategy appears to be shifting away from display ads toward an “Action Layer.” Through The Instacart integration and an Adobe partnership, OpenAI aims to facilitate high-value transactions directly within the chat interface.

This “Agentic Commerce” model would allow the company to generate revenue from economic activity—taking a cut of a grocery order or a software subscription—without cluttering the interface with traditional ads, potentially solving the revenue puzzle without alienating the user base.