Luna drops 80 percent, Terra drops a fifth, Sol holds
OpenAI cut prices for two of its three GPT-5.6 models on July 30, 2026, roughly three weeks afterGPT-5.6 reached the public. Terra, the mid-tier model, fell 20 percent to $2 per million input tokens and $12 per million output tokens. Luna, the fastest and cheapest of the three, fell 80 percent to $0.20 per million input tokens and $1.20 per million output tokens. Sol, the top-tier model, kept its price. [1][2]
OpenAI described the change as a continuation of a stated strategy: "Our strategy remains focused on advancing both capability and efficiency so each generation of intelligence can accomplish more work at a lower cost," the company said in a release. It also said Luna matches the performance of leading models from a year earlier while running a task that once cost about a dollar for roughly six cents, nearly nine times faster. All three models remain available through ChatGPT Work, Codex, and the OpenAI API. [2][1]
An efficiency story, and a market that had gotten cheaper around it
OpenAI attributed part of the savings to work its own models did on its infrastructure: it said GPT-5.6 Sol optimized GPU software on its own, cutting deployment costs by 20 percent, and that speculative decoding improved token generation by more than 15 percent. [2]
CNBC reported the cuts also came as enterprises grew less willing to deploy expensive models without a clear return on investment, and as OpenAI worked to fend off Chinese startups and other large vendors touting cheaper models. Days earlier, Moonshot AI had published the open weights ofKimi K3, which outperformed some leading American models on industry benchmarks. Anthropic had followed withClaude Opus 5, priced at half of its flagship Fable 5 while performing comparably on coding and knowledge work. Google had days before that debutedGemini 3.6 Flash, pitched as cheaper per task than Kimi K3. Microsoft's Satya Nadella separately highlighted his company's own cost-effective models on an earnings call the day before OpenAI's cut. [1]
Atlas interpretation: OpenAI's own statement credits internal efficiency work rather than competition for the cut, but the timing sits inside a two-week run of rivals underscoring price and cost-per-task in their own announcements. Neither explanation rules out the other: a model that got cheaper to run and a market getting more price-sensitive point toward the same decision, and CNBC's reporting treats the coincidence of timing as itself worth noting rather than resolving which force did more of the work. [1][2]