6 Sol 低约 30%
Key Highlights
Data from Artificial Analysis shows that GPT-6.1 Sol's Cost per Task is about $0.72, roughly 30% below GPT-6 Sol at $1.05, and GPT-6 Sol itself was already about half of GPT-5.6 Sol at $1.99. Comparing three generations of the same tier, the price-performance curve drops steeply, which is good news for anyone running models at scale and bad news for vendors charging legacy premiums.
What Happened
Cost per Task here is a blended unit price normalized to real task difficulty, which reflects what users actually pay far better than raw token prices. GPT-6.1 Sol holds a level close to Astra on coding and computer use while compressing its unit price to a little over a third of the previous generation. For buyers, the headline is not a new benchmark score but a lower bill for roughly the same work.
Technical Details
The cost drop comes from two sides: higher inference efficiency in the model itself, and a pricing strategy adjustment where OpenAI set Sol-class "value tier" token prices at roughly one-fifth of Astra. Artificial Analysis compares models on a unified task set, avoiding the inconsistency of each vendor reporting its own favorable numbers. That methodological discipline is why the $0.72 figure is credible enough to plan budgets around.
Comparison with Competitors
Within the same tier, GPT-6.1 Sol combines "good-enough flagship experience" with "clearly lower price." For price-sensitive coding and automation scenarios, it directly squeezes the room occupied by the previous Sol generation and a batch of third-party relay services. Competitors will feel pressure to match the curve or lose the cost-conscious segment that drives the highest call volumes.
Industry Impact and Use Cases
For developers, this means more steps can default to the Sol tier, with model routing escalating to Astra only when stuck. The cost structure of batch jobs, code generation, and internal tooling will be rewritten as a result. When the marginal cost of a call falls this far, product teams stop rationing intelligence and start embedding it everywhere.
Further Analysis
Put simply, models are getting cheaper faster than most people expected. Three Sol generations went from $1.99 to $0.72, cutting unit price by nearly two-thirds in a year while capability kept climbing. This dual curve of rising performance and falling price accelerates the embedding of AI into everyday software: when a single call is cheap enough to ignore, products will default to stuffing in intelligence. For startups, this is a good window to fatten margins before the next price drop resets expectations again.
Data and Methodology
The $0.72 cost per task is measured by Artificial Analysis on a unified task set, which is more credible than vendor self-reports, but the definition of a "task" still follows its methodology. Under different task difficulties, the Sol-versus-Astra value gap shifts, so do not treat one number as the truth for every scenario you run.
Risks and Limitations
The low tier may underperform flagships on very long contexts or extremely hard reasoning. Cost per Task is an average, and on tail-hard tasks Sol may retry repeatedly and end up more expensive. Before pushing cost to the absolute minimum, confirm your task distribution sits inside Sol's comfort zone.
Advice for Engineering Teams
Use a model router to set Sol as the default and Astra as the escalation tier. Re-test price-performance on real traffic periodically, and when a model gets cheaper, automatically shift more traffic to the cheaper tier so the bill keeps thinning as the technology improves.
Market Position
GPT-6.1 Sol's pricing strikes directly at the down-market: teams that want scale, margin, and AI embedded in everyday software will treat it as the default. OpenAI uses the "value tier" to grab the middle layer, protecting flagship premium while blocking competitors in the mid segment at the same time.
Extended Observation
The price war forces the whole industry to re-think pricing. When one generation's unit price drops two-thirds in a year, any team that planned its business on old prices gets embarrassed. Treating model cost as a variable cost that you reallocate dynamically as prices fall is the new normal, not a clever optimization.
Takeaway
Put simply, price cuts affect the daily bill more than new model launches. Treating Sol as the default and Astra as the escalation is the steadiest price-performance posture right now, and it is easy to wire into any routing layer you already run.