GPT-6.2 Sol lists at $2 US per million input tokens and $10 US per million output tokens. GPT-6.2 Astra lists at $8 US input and $40 US output per million tokens. Both models launched on Nov 17, 2026, in ChatGPT Work and Codex only. This rollout adds neither model to regular Chat, for any plan tier. A Canadian team that opens its usual chat window will see nothing new.
Direct answer: GPT-6.2 Astra costs $11.46 CAD per million input tokens and $57.30 CAD per million output tokens. GPT-6.2 Sol costs $2.87 CAD input and $14.33 CAD output per million tokens. Canadian prices convert the US list at US$1 = CAD 1.4326. Both models serve Plus, Pro, Enterprise and Edu users in Work and Codex. Developers reach them in the API as gpt-6.2-astra and gpt-6.2-sol.
What launched on Nov 17, 2026?
Astra is the frontier model in this pair, built for the hardest projects a team runs. Sol is the efficient model, built for everyday work and multistep agent workflows. Most Canadian firms run far more everyday work than frontier work each week. A support draft, a summary, a code fix and a research pass are Sol jobs. A hard reasoning problem with a costly error attached is an Astra job. Price the two models apart, then test each one against its own workload.
Ultrafast also changed hands. Pro 500 subscribers can use Ultrafast with either model. Pro 200 subscribers get Ultrafast for the first time, starting with GPT-6.2 Sol.
Where can a Canadian team actually use these models?
Access sits in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users. Regular Chat is outside this rollout, and that single fact will confuse many teams. Staff will hear a launch happened, open Chat, and report that nothing changed. Managers should send one short note before questions start landing in the support channel. The note names Work and Codex as the only doors into these models today.
Availability depends on plan, workspace permissions and rollout progress. Admins should update the app, restart it, and check permissions before reporting a missing model. Data handling deserves the same care as in Notion agent Memories are private to each user. Decide what they may keep.
What does GPT-6.2 cost in Canada?
| Model | Input per 1M tokens | Output per 1M tokens |
|---|---|---|
| GPT-6.2 Astra | $11.46 CAD | $57.30 CAD |
| GPT-6.2 Sol | $2.87 CAD | $14.33 CAD |
Sol costs 25% of the Astra rate on input and output alike. Send a task to Astra only when its error cost beats that price gap. A wrong client summary costs more than a few cents of extra token spend. A routine internal summary costs almost nothing to redo on a cheaper model. The token unit still misleads, a point made in Gemini 4 Carbon is out. The token price is the small number.
Cost per finished task is the number that belongs in a Canadian budget review. Astra claims new highs on professional work tests at under 1/3 of the prior cost per task. On Agents’ Last Exam, Astra at xhigh scored 5.2 points above its predecessor at max. That result came at less than 1/2 the cost per task as well. Read those figures as vendor claims, and test them on your own work.
Why does Sol change the cost per task?
On FrontierCode 2.0, Sol nearly matches Astra on merge-ready changes to real codebases. It does that work at about 1/5 of the Astra cost per task. For a Canadian software team, that claim points straight at the daily code queue. Most merge-ready changes are ordinary work, and ordinary work should carry ordinary prices. If Sol holds that result on your codebase, Astra becomes a specialist tool.
An agent that checks records and updates files spends tokens constantly, so Sol pricing changes its monthly cost shape. Astra volume needs a gate, a named owner and a reason for each run. The buying check in AI tools for Canadian small businesses, rechecked for 2026 applies here too.
What changes for teams on the API?
Both models support the Responses API and Chat Completions from the launch date. Tool calling, however, requires the Responses API under this release. A team still calling tools through Chat Completions faces a migration before it gains tools. That work belongs in the switch estimate, beside token prices and training time.
GPT-6.2 also changes how the Responses API preserves reasoning across edited conversations. When a conversation prefix changes, earlier reasoning items can drop from the request. Apps that edit history should test that behaviour before any production switch. A 2-hour background window cuts token rates by 30% in public beta. An 8-hour window cuts rates by 40%, and overnight batch work fits both windows.
What about the cyber safeguards on Astra?
Astra is described as the most capable model yet for cyber work in this line. OpenAI reports a jump of 22 points over its predecessor on an internal ExploitBench port test. Both new models launch with tighter safeguards for higher-risk cybersecurity assistance requests. A Canadian security team should expect more refusals on dual-use work at first pass. Test borderline tasks, such as audit scripts and access reviews, before rollout. The same test-first point runs through Open Google Vids before you buy Gemini Omni API.
What should a Canadian firm test before switching?
Run 10 real tasks from your own queue, split across code, documents and research. Run each task on Sol, on Astra and on the model your team uses today. Record finished tasks, failed tasks, token spend and staff review time for each run. Divide total spend by finished tasks, and compare that figure across all 3 models. No hands-on test ran for this post, so these launch claims remain vendor claims here.
The recap is short. Sol carries everyday work at $2.87 CAD input, inside Work and Codex only. Astra carries hard work at 4 times the Sol rate, with tighter cyber safeguards. Neither model sits in regular Chat, and tool calling needs the Responses API.
Run that 10-task test this week, then put 1 question to the vendor in writing: which workspace setting turns GPT-6.2 Sol on for your Canadian team, and why does regular Chat stay outside this rollout?