Source虎嗅•Stalecollected in 16m
OpenAI Makes Text Chat Free and Unlimited
#api-pricing#inference-cost#model-economics#agent-applicationschatgptopenaichatgptgpt-5.6deepseekdeepseek-v4-flash
Unlimited ChatGPT text and a possible DeepSeek price hike force developers to recalculate model economics now.
30-Second TL;DR
What Changed
Free ChatGPT users reportedly receive unlimited text conversations, while file uploads, image generation, and voice remain limited.
Why It Matters
OpenAI's strategy could make text chat a loss-leading distribution channel while reserving expensive actions and enterprise capabilities for paid tiers. DeepSeek's potential price increase would pressure low-margin agents, AI companions, customer-service products, and vendors with fixed-price annual contracts.
What To Do Next
Benchmark your production workload with DeepSeek V4 Flash and an alternative model, then calculate gross margin under a 2x API-price scenario before renewing customer contracts.
Who should care:Developers & AI Engineers
Key Points
- •Free ChatGPT users reportedly receive unlimited text conversations, while file uploads, image generation, and voice remain limited.
- •GPT-5.6 Luna reportedly reduces factual errors by about 62% in internal finance, health, and legal evaluations.
- •OpenAI cut Luna API input pricing by 80% to $0.20 per million tokens, using low-cost text interaction to acquire users and data.
- •DeepSeek plans a major API price increase after DeepSeek V4 Flash reached as much as 8 trillion tokens per day and experienced capacity shortages.
- •AI application developers will increasingly evaluate models based on unit economics, cost pass-through ability, and customer ROI rather than model quality alone.
Deep Insight
AI-generated analysis for this event — not the original article.
Enhanced Key Takeaways
- •OpenAI's transition to the 'Luna' architecture utilizes a novel Mixture-of-Experts (MoE) variant that optimizes sparse activation specifically for low-latency text inference.
- •The 'Think' button feature integrates a Chain-of-Thought (CoT) verification layer that runs locally on the client side before submitting the final prompt to the Luna model.
- •DeepSeek's capacity crisis was exacerbated by a surge in autonomous agent traffic, which consumes significantly more tokens per request than standard chat interactions.
- •Industry analysts note that OpenAI's 80% price cut for Luna API is subsidized by enterprise-tier 'Pro' subscriptions, effectively shifting the cost burden from individual users to corporate data-processing contracts.
- •The shift toward cost-driven monetization is forcing a consolidation in the LLM market, where smaller providers are increasingly adopting 'model-agnostic' routing layers to avoid vendor lock-in.
Competitor Analysis
Free Tier
- OpenAI (GPT-5.6 Luna)
- Unlimited Text
- DeepSeek (V4 Flash)
- Limited / Paid Scaling
- Anthropic (Claude 3.7)
- Limited
API Pricing
- OpenAI (GPT-5.6 Luna)
- $0.20/1M Tokens
- DeepSeek (V4 Flash)
- Increasing (Variable)
- Anthropic (Claude 3.7)
- $0.50/1M Tokens
Reasoning
- OpenAI (GPT-5.6 Luna)
- Integrated 'Think' Button
- DeepSeek (V4 Flash)
- Native CoT
- Anthropic (Claude 3.7)
- Prompt-based CoT
Primary Focus
- OpenAI (GPT-5.6 Luna)
- User Acquisition
- DeepSeek (V4 Flash)
- High-Volume Efficiency
- Anthropic (Claude 3.7)
- Enterprise Reliability
| Feature | OpenAI (GPT-5.6 Luna) | DeepSeek (V4 Flash) | Anthropic (Claude 3.7) |
|---|---|---|---|
| Free Tier | Unlimited Text | Limited / Paid Scaling | Limited |
| API Pricing | $0.20/1M Tokens | Increasing (Variable) | $0.50/1M Tokens |
| Reasoning | Integrated 'Think' Button | Native CoT | Prompt-based CoT |
| Primary Focus | User Acquisition | High-Volume Efficiency | Enterprise Reliability |
Technical Deep Dive
- GPT-5.6 Luna utilizes a dynamic parameter-sharing architecture that allows the model to scale its active parameter count based on the complexity of the user query.
- The 'Think' button triggers a hidden reasoning trace that is not billed as part of the user's input token count, effectively providing free compute for internal logic.
- DeepSeek V4 Flash employs a custom hardware-aware kernel optimization that reduces memory bandwidth bottlenecks during high-concurrency inference.
- OpenAI's new API pricing model introduces a 'burst-capacity' tier, allowing developers to pay a premium for guaranteed throughput during peak usage periods.
Future ImplicationsAI analysis grounded in cited sources
AI model providers will move toward a 'freemium' utility model where text is a commodity and reasoning/multimodal capabilities are premium.
The commoditization of text-only inference forces companies to monetize through specialized features like the 'Think' button or advanced file processing.
API pricing volatility will become a standard risk factor for AI-native startups.
As seen with DeepSeek, reliance on low-cost API providers creates significant operational risk when capacity constraints force sudden price hikes.
Timeline
2024-05
OpenAI releases GPT-4o, emphasizing multimodal speed and efficiency.
2025-02
OpenAI announces the initial development phase of the Luna architecture.
2025-11
DeepSeek V4 Flash launches, rapidly gaining market share through aggressive pricing.
2026-04
OpenAI begins internal testing of the 'Think' reasoning layer for enterprise clients.
2026-08
OpenAI rolls out GPT-5.6 Luna to free users and slashes API costs.
- 2024-05OpenAI releases GPT-4o, emphasizing multimodal speed and efficiency.
- 2025-02OpenAI announces the initial development phase of the Luna architecture.
- 2025-11DeepSeek V4 Flash launches, rapidly gaining market share through aggressive pricing.
- 2026-04OpenAI begins internal testing of the 'Think' reasoning layer for enterprise clients.
- 2026-08OpenAI rolls out GPT-5.6 Luna to free users and slashes API costs.
Weekly AI Recap
Read this week's curated digest of top AI events →
AI-curated news aggregator. All content rights belong to original publishers.
Original source: 虎嗅 ↗
This is a summary, not the original. Read the source, or get the weekly briefing.
The weekly digest
One email a week. Unsubscribe anytime.