According to CNA and Inside.com.tw reports, OpenAI announced on July 30, 2026 that it would cut API pricing for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20%, only 21 days after the three-model GPT-5.6 series launched on July 9. Flagship model Sol kept its $5/$30 per-million-token pricing unchanged.
What was cut, when, and across which models?
According to CNA, OpenAI announced on July 30, 2026 that it would reduce prices for its GPT-5.6 model family, with GPT-5.6 Terra down 20% and GPT-5.6 Luna down 80% (E1). CNA also confirmed the GPT-5.6 series — comprising flagship Sol, balanced mid-tier Terra, and cost-efficient Luna — had launched only three weeks earlier (E2).
Inside.com.tw pinned the exact dates: the three models went live on July 9, 2026, and the price cut was announced on July 30, 2026 — a gap of 21 days (E11). Sam Altman confirmed the same figures directly on social media, posting "80% drop for GPT-5.6 Luna... 20% drop for GPT-5.6 Terra," while also noting Sol was gaining a new "Fast mode" in the API at 2.5x the speed for 2x the price with unchanged intelligence (E12). Sol's base pricing itself was not cut.
What do the new API prices look like model by model?
CNA reported the post-cut rates: Terra now costs $2 per million input tokens and $12 per million output tokens; Luna costs $0.20 per million input tokens and $1.20 per million output tokens; Sol's price was left unchanged (E3). Cnyes.com corroborated the identical figures for Terra and Luna (E7).
Inside.com.tw added the before-and-after comparison: Luna's input price fell from $1 to $0.2 and its output price fell from $6 to $1.2; Terra moved from $2.5/$15 to $2/$12; Sol remained frozen at $5/$30, unchanged from the prior GPT-5.5 generation (E13). The same outlet noted that OpenAI's own July 29 engineering blog post had described Luna as "80% cheaper than Sol" — but one day later, after the cut, that gap widened to 96% cheaper, pushing the family's internal price spread from roughly 5x to 25x between Luna and Sol (E14).
| Model | Input (pre-cut) | Output (pre-cut) | Input (post-cut) | Output (post-cut) | Change |
|---|
| Luna | $1.00 | $6.00 | $0.20 | $1.20 | -80% |
| Terra | $2.50 | $15.00 | $2.00 | $12.00 | -20% |
| Sol | $5.00 | $30.00 | $5.00 | $30.00 | unchanged |
Source: E13, cross-checked against E3, E7, E12.
How does the new pricing stack up against Anthropic, Google, and Chinese rivals?
Inside.com.tw compared the new Luna rate to Anthropic's cheapest model: Claude Haiku 4.5 is priced at $1/$5 per million tokens, meaning the post-cut Luna input price is only one-fifth of Haiku 4.5's input rate. At the flagship tier, however, the comparison flips — Sol's $30 output price is higher than Claude Opus 5's $25 output price (E17).
The same outlet cited external reporting on Chinese rival Zhipu AI's GLM-5.2, priced at roughly $1.4/$4.4 per million tokens — putting its input price at about three-tenths of Sol's and its output price at under half of Sol's (E19). Separately, per Cnyes.com, Google introduced several cost-focused models this month, claiming its Gemini 3.6 Flash has a lower per-task cost than Kimi K3, though no specific figures were disclosed for that comparison (E9).
What competitive pressure is OpenAI responding to?
Cnyes.com reported that Chinese startup Moonshot AI released its open-weight Kimi K3 model in mid-July, which reportedly surpassed leading U.S. models on some industry benchmarks and drew attention in Silicon Valley (E4). In the wake of that release, Anthropic launched Claude Opus 5, which the outlet said performs close to a higher-tier product on coding and knowledge-work tasks while costing only half as much (E8). Google, per the same report, rolled out multiple cost-oriented models this month, including Gemini 3.6 Flash (E9), and Microsoft CEO Satya Nadella reportedly emphasized the importance of cost-effective AI models on a recent earnings call (E10).
Inside.com.tw, citing CNBC, framed the pressure as coming from two directions: enterprises increasingly scrutinizing AI return-on-investment and showing little appetite for deploying high-priced models, and rivals — Chinese AI startups alongside Google and Microsoft — all pushing high-value-for-money models (E18).
What technical justification did OpenAI give for the price cut?
According to Inside.com.tw, OpenAI's July 29 engineering blog post said GPT-5.6 Sol had autonomously rewritten production GPU kernel code via Codex, and combined with broader kernel improvements, this cut end-to-end serving costs by 20%. Sol also reportedly ran hundreds of experiments improving its speculative-decoding draft model, lifting token-generation efficiency by more than 15% (E15).
The same source flagged an important caveat: these are OpenAI's own internal figures, not independently verified by a third party — though the outlet noted the direction is clear, with the model itself reportedly taking on work aimed at lowering its own serving costs (E16).
How does the launch-to-cut timeline compare with the market reaction?
CNA confirmed the GPT-5.6 series — Sol, Terra, and Luna — launched three weeks before the July 30 price announcement (E2). Cnyes.com dated the launch precisely to July 9, 2026 (E6). Inside.com.tw calculated the exact interval between the July 9 launch and the July 30 cut as 21 days (E11). Within that same window, per Cnyes.com, Moonshot AI's Kimi K3 launched in mid-July, followed by Anthropic's Claude Opus 5 and Google's cost-focused Gemini 3.6 Flash release, both cited by the outlet as market responses during the same period (E4, E8, E9).
What this means
The timeline itself is the clearest signal in the evidence: OpenAI's July 29 blog post described Luna as 80% cheaper than Sol, and the very next day the actual price cut pushed that gap to 96% (E14) — a reversal that happened one day apart. That 21-day span between launch and repricing (E11) overlaps precisely with the period in which Moonshot AI's Kimi K3, Anthropic's Claude Opus 5, and Google's Gemini 3.6 Flash all entered the market (E4, E8, E9). Yet the new pricing is not uniformly aggressive: Luna undercuts Anthropic's Haiku 4.5 by a factor of five on input tokens (E17), but Sol's $30 output price still sits above Claude Opus 5's $25 (E17) and well above Zhipu AI's GLM-5.2 output rate of $4.4 (E19). OpenAI attributes part of the savings to a 20% reduction in serving costs and a 15%-plus efficiency gain from Sol's own kernel and decoding optimizations (E15) — figures the company itself acknowledges are internal and not third-party verified (E16).