AI Daily · Nov 18, 2026: OpenAI launches GPT-6.2 Astra and Sol with 1.05M context and steep cost cuts
OpenAI's GPT-6.2 Astra and Sol bring a 1.05M-token context and adjustable reasoning effort, with Astra at $8/$40 per million tokens and under a third of GPT-6 Astra's cost per task. Tencent also open-weights the 224B Hy4 flash.
Pick a date
-
OpenAI GPT-6.2 Astra and Sol Go Live With 1.05M Context
OpenAI released GPT-6.2 Astra and GPT-6.2 Sol reasoning models with a 1.05M token context window, adjustable reasoning effort from low to max, multimodal image understanding, computer use and multi-agent delegation; Astra targets hard software engineering while Sol is the cheaper option for everyday coding and agent workflows, available via API and web chat.
-
GPT-6.2 Astra and Sol go live on OpenRouter with per-token pricing
On OpenRouter, GPT-6.2 Astra is priced at $8/M input and $40/M output tokens, while GPT-6.2 Sol costs $2/M input and $10/M output.
-
GPT-6.2 Astra and Sol roll out in ChatGPT Work and Codex
Both models are available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, with Ultrafast extended to Pro 200 subscribers starting with GPT-6.2 Sol; developers get gpt-6.2-astra and gpt-6.2-sol in the OpenAI API.
-
Tencent open-weights Hy4 flash with 224B parameters and 1M context
Tencent released Hy4 flash as open weights, including Hy4 flash-FP8 and Hy4 flash-Base: 224B total parameters with 12B active, 1M context and native multimodality. Tencent says it matches Claude Haiku 5.5 (high) on daily-work blind tests at an 18x lower task cost, with API cache pricing from $0.003 per 1M tokens.
-
Perplexity launches Orchestrator API routing across 85+ models
Perplexity's new Orchestrator API puts open-weight and proprietary models behind one endpoint with unified billing, auto-selecting providers by availability, capacity and latency. It runs on the same layer that routes hundreds of millions of production requests across 85+ models and a dozen-plus providers; each model is billed at its published per-token rate, with spend tracking by model, project
-
White House and Treasury weigh public disclosure of Chinese open-weight model use
Per Reuters, the White House and Treasury are discussing a proposal requiring US companies to publicly disclose on their websites their use of Chinese open-weight AI models, aiming to discourage adoption via public pressure.
-
B.AI cuts DeepSeek-V4.1-Flash discount to 50% off, lists other model deals
B.AI raised the DeepSeek-V4.1-Flash discount from 40% to 50% off; the 763B open model supports 1M context and interleaved image-text input. Other deals: HY4/HY4-Flash 50% off (until Dec 31 23:59 SGT), GLM-5.3-Flash 40% off, Qwen3.8-Flash and MiniMax-H3-Image 20% off, Qwen4-Max 10% off.
-
OpenCode offers $200 of GLM 5.3 usage for $10
OpenCode launched phase 2 of its GLM promotion: $200 of GLM 5.3 usage for $10, available to all OpenCode Go subscribers, up from the earlier $100-for-$10 offer.
-
OpenRouter management keys gain read-only and workspace-scoped access controls
OpenRouter management keys can now be made read-only, restricted to a single workspace, or both, limiting the impact of a leaked key; available today on all plans.
-
Qwen3.8-27B LoRA adaptation lifts coding agent benchmarks
A LoRA adapter trained with reinforcement learning adapts Qwen3.8-27B for the pi coding agent, raising SWE-bench Pro (Hard 51) pass@1 from 29.4% to 62.7% and Terminal-Bench 2.1 from 39.5% to 55.8%, while cutting tool calls per task by 55–62%.