OpenAI has released GPT-6.2 Astra and GPT-6.2 Sol, two new models now live in ChatGPT Work, Codex and the API, with Astra priced at $8 per million input tokens and $40 per million output tokens, and Sol at $2 and $10. In the same week Google launched Gemini 4 Neon at $3 per million input tokens and $15 per million output, with a one million token context window.
Both companies pitched these as their best models for serious professional work. For anyone who pays for AI by the token, whether a freelance developer billing clients or a small agency running chatbots for local businesses, the more useful question is which one delivers the work at a price that still leaves a margin.
What OpenAI shipped
According to OpenAI’s announcement, Astra is the frontier model aimed at hard projects: complex coding, computer use, 3D and creative work, and science. Sol is the efficient everyday model and becomes the default for routine and agent-style tasks.
The pricing details matter more than the headline. Astra’s input and output rates are 20% lower than the previous GPT-6 Astra, and cached input drops to $0.40 per million tokens. Sol keeps the rates of GPT-6.1 Sol: $2 input, $0.10 cached input and $10 output per million tokens.
OpenAI says Astra scores 5.2 points higher than GPT-6 Astra on the Agents’ Last Exam benchmark, and on its internal GDPval v2 test of real-world work tasks, Astra gained 6.2 points and Sol 5.3 points over their predecessors. One figure stands out for budget-conscious users: on Cognition’s FrontierCode 2.0 coding test, OpenAI reports Sol landing within 0.5 points of Astra. If that holds in your own work, Sol does most coding jobs at a quarter of Astra’s price.
Inside ChatGPT, both models are available to Plus, Pro, Enterprise and Edu users. OpenAI’s page makes no mention of free-tier access.
Google’s answer: Gemini 4 Neon
Google’s Gemini 4 Neon launch post positions the model for complex coding, analytical reasoning and production business workflows. Cached input costs $0.15 per million tokens, a 95% discount on the standard rate, and the context window reaches one million tokens, enough to load a large codebase or a stack of long documents in one request.
Google cites a score of 61 on the Artificial Analysis Intelligence Index, which it says ranks third overall, and a first-place 58.6 on the Artificial Analysis Coding Agent Index. Developers can use it through the paid tier of the Gemini API in AI Studio, and Google AI Pro and Ultra subscribers can pick it in the Gemini app.
What the same job costs on each model
Benchmarks are self-reported, so the cleanest comparison is the published price list. Take a simple workload: one million input tokens and 200,000 output tokens, roughly what a freelancer might spend summarising a month of client documents or running a busy support bot. Using each company’s list prices, before caching:
- GPT-6.2 Astra: $8 input plus $8 output, about $16.
- Gemini 4 Neon: $3 input plus $3 output, about $6.
- GPT-6.2 Sol: $2 input plus $2 output, about $4.
Caching changes the picture further when you send the same instructions or reference files again and again. A support bot with a long fixed system prompt pays the cached rate on that repeated portion, which is where Google’s 95% cache discount and OpenAI’s $0.10 Sol cached rate can cut real bills sharply.
Which one should a small team reach for?
For freelancers and small agencies in Pakistan and the Gulf, where every dollar of API cost comes out of a fixed client fee, a tiered setup makes sense. Route routine work like drafting, classification and simple fixes to the cheapest model that passes your own quality checks, and save the premium tier for the hard tasks where a better answer saves hours. Microsoft added to the pressure this week too, launching MAI Code 2 and MAI Code 2 Flash inside GitHub Copilot as lower-cost coding options.
The bigger shift is in how AI work gets priced to clients. Beingguru’s look at how AI freelancing pays through bigger projects rather than higher hourly rates found that the money is moving toward larger, outcome-based jobs. Cheaper capable models feed that trend: when the underlying cost of a task falls, the freelancer who quotes per project keeps the difference, while the one who bills by the hour hands it back to the client.
Prices in this market move every few months. Before committing a client project to any one model, run a small test batch of your real tasks through two or three options, record the cost and the quality, and let your own numbers decide.





