以 Astra 五分之一价格接近其编码与计算机操作水平
Key Highlights
OpenAI released GPT-6.1 Sol at $2 per million input tokens, $10 per million output, and $0.10 per million cached input—roughly one-fifth of GPT-6 Astra's standard price. The pitch is direct: get close to Astra's coding and computer-use level at about a fifth of its cost. For price-sensitive, high-volume use, this is the cut that brings frontier capability inside the budget instead of leaving it as a line item teams delete. The pricing itself is the headline, not a benchmark score.
What Happened
Sol is OpenAI's "value" tier alongside Astra. It does not chase the absolute top of the chart; instead it compresses the high-frequency abilities—coding and agentic computer use—into a price that scales. The $0.10 cached input is the quietly important number: in long-context, multi-turn tasks, repeated context riding the cache dramatically amortizes cost, which means a "long session" no longer equals "burning money" the way it did on per-token-full-price models a year ago.
Technical Details
The pricing structure reveals the positioning: cheap input, expensive output, matching enterprise usage that reads far more than it writes; near-free cached input rewards keeping system prompts, knowledge bases, and long documents resident in context. This billing design nudges developers to front-load reusable content into the cache instead of resending it every turn, directly cutting the bill. For teams doing RAG or long-document analysis, Sol's unit economics are materially better than a full-price flagship with no cache discount.
Comparison with Competitors
Against the value tiers of Claude and Gemini, Sol's killer is "near-Astra ability for a fifth of the price." Against pure small models, it keeps a safety net for complex coding tasks. For budget-limited teams, this is the "20 percent of the money for 80 percent of the effect" trade that is usually cheaper than clinging to the smallest model and far more sustainable than a full-price flagship. The realistic comparison is total cost at your own traffic, not a leaderboard.
Industry Impact and Use Cases
For startups and internal-tool teams, Sol turns "give every employee a coding assistant" from a luxury into a line item that survives a budget review. For AI product companies, it lowers inference cost, helping margins or enabling price wars. Model pricing is shifting from flagship premiums toward tiered coverage, and Sol is the representative of that trend and will push rivals to rethink their own value tiers sooner rather than later.
Further Analysis
The cache pricing is the part worth copying in your own architecture: design prompts and retrieval so the expensive, stable context is cached and the cheap, variable part is what you pay per turn for. Teams that do this can run Sol at a fraction of the naive cost and still get Astra-class results on the hard steps. Treat Sol as the default workhorse and reserve Astra for the handful of jobs where the extra capability clearly pays for itself, and your blended cost drops without a visible quality hit.
Further Analysis
Adoption should start with the highest-volume, lowest-stakes coding tasks—boilerplate, refactors, test scaffolding—where Sol's near-flagship level is more than enough and the savings compound across thousands of calls a day. Measure acceptance rate and rework, not just latency, because the value is in fewer human fixes, not faster first drafts. As usage grows, the cached-input discount does the heavy lifting, so the marginal cost per task stays flat even as context grows, which is what makes "give everyone an agent" affordable rather than aspirational.
Further Analysis
The strategic read is that OpenAI is consciously building a pricing ladder, not a single flagship, which mirrors how cloud compute went from one big machine to granular tiers. That pressures every lab to offer a credible mid tier or lose the volume segment to Sol. For buyers, the lesson is to stop benchmarking models in isolation and start benchmarking cost-per-good-outcome at your own scale, because that is where Sol wins and where the next year of competition will actually be fought.
Market Position
GPT-6.1 Sol's play is to grab scale with a value tier: teams that embed AI in everyday software and run batch tasks will treat it as the default. OpenAI keeps the Astra flagship premium while using Sol to block the mid segment, squeezing both competitors and relay services that lived on the price gap nobody noticed closing.
Extended Observation
Three Sol generations cut unit price nearly two-thirds in a year while capability kept climbing. This curve, where price cuts affect the bill more than new launches, forces the whole industry to re-price and gives small teams real confidence to stuff intelligence into products without fearing the invoice that arrives at month end.