OpenAI's advertised 50% GPT-6 Sol price cut is built on a retired model tier.
The AMW Read
Debunks OpenAI's headline price cut as a retired-tier relabel while confirming Anthropic's genuine cache-cost engineering and DeepSeek's architecture edge, updating the segment's pricing-war narrative without overturning a top-tier claim.
OpenAI's advertised 50% GPT-6 Sol price cut is built on a retired model tier.
OpenAI launched GPT-6 Sol and Luna the same day Anthropic shipped Claude Opus 5.5. Sol now costs $2 per million input tokens and $10 per million output tokens, nominally half of GPT-5.6 Sol. But OpenAI's three-tier lineup — flagship Sol, mid-tier Terra, light Luna — quietly became Astra, Sol, Luna, dropping Terra. GPT-5.6 Terra, priced at $2 input and $12 output per million tokens, is nearly identical to the new "discounted" Sol: last generation's mid-tier model moved up a rung and got relabeled. Anthropic's Opus 5.5 cut list price 40% versus Opus 5 and cut cached-token read costs 60%.
The split matters because it separates real engineering-led cost cuts from commercial repositioning inside the same price war. DeepSeek's V4.1-Flash still holds a structural edge, activating only 8-16B of 552B total parameters with a KV cache near 890 bytes per token. OpenAI's gains lean more on parameter distillation into lighter MoE branches plus commercial levers: a 90% discount on cache-hit input tokens (to $0.20 per million) and a pricing cliff beyond 272K tokens that doubles input cost. That cache discount also builds switching-cost lock-in for enterprises anchoring system prompts in OpenAI's infrastructure. Per the AI Market Watch index, OpenAI logged 336 tracked news items in the past 90 days versus 297 prior — a count limited to pipeline-ingested sources — underscoring how fast pricing claims now face scrutiny, echoing the recent pushback on its ARC-AGI-3 benchmark claim.
For builders, sub-272K-token workloads and cache-hit rate matter more than sticker price when comparing Sol, Opus 5.5, and V4.1-Flash; one cited enterprise deployment raised cache-hit rates from 85% to above 90%, moving effective cost more than any list-price change. For investors, DeepSeek's cost edge over frontier US labs is narrowing but not closed, and the segment's real differentiator is shifting from base token price to caching architecture and lock-in mechanics.
