API pricing for GPT-6.2 Astra sits 20 percent below its predecessor's at $8 per million input tokens, $0.40 for cached inputs, and $40 per million output tokens. OpenAI has released GPT-6.2 Astra and GPT-6.2 Sol across Plus, Pro, Enterprise, and Edu tiers. Astra targets specialized workflows in science, law, and finance, while Sol handles everyday operational tasks. Enterprise budget reallocation is accelerating as corporate buyers shift capital away from legacy infrastructure toward these high-efficiency models, forcing internal IT departments to restructure deployment pipelines to capture lower per-token costs. As companies migrate multi-agent architectures to the new endpoints, infrastructure providers face intense operational cost pressures to match these price cuts without destabilizing their operating margins.
OpenAI is driving down the cost of advanced computational workflows, shifting enterprise competition directly toward the unit economics of deployment. The central question for corporate buyers is whether Anthropic can protect its enterprise footprint without compressing its margins under OpenAI's pricing structure. The downward cost trend is accelerating: the Sol tier delivers capabilities close to flagship models, but at a fraction of the inference spend.
Benchmark Performance and Cost Disruption
On the legal benchmark Harvey LAB-AA, which measures contract clause analysis, OpenAI reports that GPT-6.2 Astra more than doubles the pass rate of Anthropic's Claude Fable 5.2 at roughly one-sixth of the cost. On Agents' Last Exam, a benchmark tracking multi-step professional workflows, GPT-6.2 Astra scores 5.2 points above its predecessor GPT-6 Astra at less than a third of the cost, while leading Claude Fable 5.2 by 5.5 points at 5 percent of the cost. Only Google's Gemini 4 Carbon matches Astra on this benchmark.
According to Artificial Analysis, it dominates the Pareto frontier in terms of cost-performance ratio.
On the internal GDPval-v2 benchmark covering consulting, finance, and design, OpenAI states that Astra gains 6.2 percentage points over its predecessor while using 22 percent fewer output tokens. OpenAI designates Astra as its primary model for scientific workloads alongside automated slide generation structured to pull layout and typography rules from templates. On Terminal-Bench-Science, which tests terminal-based scientific workflows, the model scores 9 percentage points above GPT-6 Astra at less than a third of the cost. Deployment case studies from early enterprise adopters indicate that these efficiency gains directly shorten project delivery timelines for complex financial modeling and automated contract review.
The Efficiency Frontier and Operational Limits
Anthropic's models maintain an anchor at the top of raw capability rankings. In the Artificial Analysis Intelligence Index, GPT-6.2 Astra ranks second behind Claude Fable 5.2. On the 3D benchmark AA-3D, where human evaluators compare blinded interactive experiences, GPT-6.2 Astra leads with 1,318 Elo points, ahead of Claude Fable 5.2 at 1,271 and GPT-6.2 Sol at 1,259.
Efficiency gains bring operational friction. Both GPT-6.2 Astra and Sol cross the critical cybersecurity threshold in OpenAI's Preparedness Framework. While OpenAI plans to distribute the new models through OpenAI Daybreak to vetted security researchers with fewer guardrails, it is implementing stricter safeguards on standard interfaces, which OpenAI acknowledges will restrict independent vulnerability research and security testing.
Everyday execution exhibits parallel margin compression. GPT-6.2 Sol maintains GPT-6.1 Sol's pricing at $2 for input, $0.10 for cached inputs, and $10 for output per million tokens, while trailing Astra by 0.5 points on the coding benchmark FrontierCode 2.0 at one-fifth of the cost.
Against the field
By dropping the entry price to eight dollars per million input tokens, OpenAI deliberately drags the platform rivalry away from raw capability scores into ruthless per-token accounting. In early October the provider already delivered near-flagship intelligence at a fifth of the price, and it is now forcing competing infrastructure vendors to sacrifice their own operating margins just to keep enterprise workloads. Can Anthropic defend its enterprise accounts against Astra without resorting to a margin-crushing price cut on Claude Fable 5.2?
Our stories: OpenAI Sol: Why Logic Trumps Speed in Business AI (June 30); GPT 6.1 Sol Cuts Frontier AI Inference Costs by 80% (October 6).