Improving GPT‑5.6 Sol in ChatGPT, expanding GPT‑5.6 Luna access for free users
OpenAI’s move to make its new GPT‑5.6 “Luna” model with optional reasoning available for free in ChatGPT is seen as both a major capability upgrade for casual users and a sign of intensifying price and performance competition in the AI market. Commenters debate whether this actually downgrades occasional access to higher-end models, criticize confusing model and “reasoning level” choices, and question how sustainable generous free tiers are as foundation models become commoditized. The changes also revive arguments over whether current systems already qualify as “AGI,” how they will be monetized (ads, data, integrations), and what widespread reliance on such tools might mean for culture and work.
Luna Free Tier & Overall Change
- Many see “free unlimited Luna + Think button” as a major upgrade over prior weak free models (e.g., 5.5 Instant / mini with tight caps).
- Some argue it’s a downgrade vs. the brief burst of stronger models free users previously got (5.x then auto‑downgrade), since Luna is described as “nano tier.”
- Several suspect Luna’s efficiency has improved enough that generous free usage now makes economic sense and yields more training data and usage metrics.
Model Quality: Luna, Terra, Sol and Competitors
- Opinions diverge sharply:
- Some say Terra is “awful” and recommend only Luna Max or Sol Medium/High.
- Others find Terra strong via Copilot and “leaps ahead” of some rival models.
- Luna is widely seen as surprisingly capable for its price, but with less knowledge and “small model” feel; best for agentic/web tasks.
- Sol High/XHigh is the go‑to for complex work; Ultra/Max seen as costly and overkill except when others fail.
- Comparisons with external models (DeepSeek, Kimi, Gemini, Claude, Grok) show a crowded, rapidly commoditizing landscape.
Reasoning Modes, Sliders & UX Friction
- Many dislike the proliferation of effort levels (Instant/Low/Medium/High/XHigh/Max, Pro/Ultra) and find it confusing or “batshit crazy.”
- Some want the system to auto‑choose or ask clarifying questions instead of exposing knobs; others appreciate manual control for long, deep tasks.
- Concerns that users default to “high” out of fear of errors, potentially increasing token usage; some suspect dark‑pattern routing toward cheaper modes.
- Confusion about multiple “versions” of Sol (Chat vs Work/Codex) under the same name.
Business Model, Competition & Ads
- Thread notes intense commoditization pressure; bottom of the market perceived as collapsing into free.
- Speculation that free Luna targets Google’s AI Overviews and other generous free tiers.
- Debate over whether Google’s vertical integration and custom hardware will dominate the low‑end, despite current model quality complaints.
- Some expect monetization through profiling and ads; others question whether ad economics work at LLM inference costs.
AGI Terminology and Capabilities
- The mission statement mentioning AGI sparks debate:
- Some think current models reasonably qualify as AGI (general, widely useful).
- Others insist they are still too error‑prone, lack embodied understanding, and fail strong Turing‑test‑like standards.
- Discussion notes “jagged intelligence”: excellent on some benchmarks (math, coding, IQ tests) yet bizarrely weak in other areas.
Free vs Paid Experience and Limits
- Paid users complain the default chat often silently uses the same cheap “Instant” model as free, with effort sliders hidden or reset.
- Some see this as a dark pattern; others as latency‑oriented design.
- Power users emphasize the real value in paid plans is coding agents/API (Codex/Claude Code) rather than chat limits.