November 17 agent guide: GPT-6.2 Astra and Sol
November 17, 2026 reset OpenAI’s professional-work stack. OpenAI introduced GPT-6.2 Astra for hard multi-step projects and GPT-6.2 Sol for everyday agentic work, with lower token prices on Astra and Critical cybersecurity ratings on both.

In this post (7 sections)
November 17, 2026 opens the next Bacancy Technology window after November 16’s Qwen4-Max brief. OpenAI shipped GPT-6.2 Astra and GPT-6.2 Sol as a professional-work pair: higher quality on long agent tasks, fewer output tokens, and a 20% Astra price cut versus GPT-6 Astra.
What shipped
| Model | Role | API price (per 1M tokens) | Builder action |
|---|---|---|---|
| GPT-6.2 Astra | Frontier hard projects, science, CAD, legal, slides | $8 / $0.40 cached / $40 | Re-benchmark against Fable 5.2 and Gemini 4 Carbon on merge and office agents |
| GPT-6.2 Sol | Everyday agentic and knowledge work | $2 / $0.10 cached / $10 | Default lane for high-volume coding and LifeVal-style tasks |
Why Astra matters for agents
OpenAI positions GPT-6.2 Astra for multistep professional workflows: slides and templates that match field conventions, reconciling workbooks, legal deliverables, computer use in CAD and simulation, and cleaner frontend and 3D work. On Agents’ Last Exam, OpenAI reports Astra 5.2 points above GPT-6 Astra at less than a third of the cost, and 5.5 points ahead of Claude Fable 5.2 at about 5% of the cost. Terminal-Bench-Science 0.1 and the Vals Science Index are cited as new highs versus GPT-6 Astra.
Sol as the volume lane
GPT-6.2 Sol keeps GPT-6.1 Sol pricing while claiming stronger multistep agentic workflows and clearer writing. On Cognition’s FrontierCode 2.0, OpenAI says Sol comes within 0.5 points of Astra at about one-fifth the cost at max effort. That is the routing signal: put merge-bar work on Astra only when Sol fails the harness.
Safety and Daybreak
Both models meet Critical cybersecurity and High biological/chemical thresholds under OpenAI’s Preparedness Framework. Astra is also OpenAI’s first external release to meet High for AI self-improvement. Launch safeguards restrict some higher-risk cyber assistance; OpenAI points approved defenders to Daybreak. Bacancy Technology’s rule matches the November 5 GPT-6 Cyber brief: keep Critical cyber capability inside authorized programs.
Builder checklist
- 01Add gpt-6.2-astra and gpt-6.2-sol to the routing boardKeep Fable 5.2 and Gemini 4 family in the same eval set; do not swap production defaults on launch charts alone.
- 02Measure tokens per completed taskOpenAI’s pitch is quality plus fewer output tokens. Track cost per merge and per office deliverable, not only pass rate.
- 03Review cyber and bio allowlistsExpect tighter refusals on dual-use prompts. Route authorized security testing through Daybreak-approved paths.
- 04Enable Ultrafast where seats qualifyPro 500 keeps Ultrafast on both models; Pro 200 gains Ultrafast starting with Sol.
Conclusion
November 17 is an OpenAI routing day. GPT-6.2 Astra raises the professional-agent ceiling at lower Astra token prices. GPT-6.2 Sol remains the volume default. Bacancy Technology’s working rule: Sol first, Astra for failures, Critical cyber behind Daybreak gates.
Related reading
- November 16 Qwen4-Max GA: /blog/november-16-qwen4-max-claude-code-326-2026
- November 14 Coding Agent Index: /blog/november-14-cursor-security-coding-agent-index-2026
- November 9 Fable 5.2: /blog/november-09-claude-fable-5-2-muse-ember-2026
- November 5 GPT-6 Cyber: /blog/november-05-gpt6-cyber-cursor-live-apps-2026
- Agentic AI topic map: /topics/agentic-ai
Need GPT-6.2 routing and Daybreak controls?
Bacancy Technology designs model routing boards, usage-limit plans, and authorized cyber-agent programs. Consulting: /consulting/ . Contact: /contact/ .
Sources: OpenAI .
Agentic AI patterns, delivered Thursdays
What I am shipping, watching, and pruning out of client stacks each week. One email. No fluff.