Updated August 23, 2026. This review focuses on information that technology owners, developers, and procurement teams can verify before testing a model or changing a production setting. It is not a general model ranking, because company-published benchmark results are not enough for a direct comparison by themselves.
Quick summary
The clearest general-purpose releases reviewed this month are OpenAI’s August GPT-5.6 Sol and GPT-5.6 Luna updates for ChatGPT, and Google’s generally available Gemini 3.7 Flash. OpenAI announced the August versions on August 6, 2026; they are distinct from July versions still used in Codex and ChatGPT Work. Google’s documentation describes Gemini 3.7 Flash as generally available, with a 1M-token context window, a 64K maximum output, and low, medium, and high thinking levels.
This article does not recommend one model as a universal default. Before making a change, record and explicitly configure the exact model ID or version used by each deployment, then verify its lifecycle and pricing status. The Arabic/English paired prompts, tool-call tests, and small regression suite described below are this article’s editorial test plan, not results established by Gemini documentation.
OpenAI: the August GPT-5.6 versions
According to OpenAI’s deployment-safety update, the updated GPT-5.6 Sol and GPT-5.6 Luna versions were released for ChatGPT on August 6, 2026. The page says these August versions are distinct from July versions still used in Codex and ChatGPT Work. Teams should therefore not assume that the GPT-5.6 family name alone identifies the same behavior or lifecycle across products. See OpenAI’s August GPT-5.6 update.
OpenAI classifies the August GPT-5.6 Sol and Luna versions as High capability in cybersecurity and biological/chemical domains, while placing them below its Critical threshold. These are OpenAI’s Preparedness Framework designations, not independent risk assessments. Review the classifications in OpenAI’s official update.
In an August 21 update, OpenAI said it reduced GPT-5.6 Sol API and credit pricing by more than 20% for three months. Do not treat that temporary reduction as a permanent price assumption; record when your use begins and recheck pricing before expanding a deployment. See the temporary reduction in OpenAI’s August 21 update.
Google: Gemini 3.7 Flash
Google announced Gemini 3.7 Flash on August 13, 2026, describing improvements in coding, knowledge work, web development, and agentic workflows. Those are the company’s descriptions of the release, not the result of independent testing in this article. Read Google’s Gemini 3.7 Flash announcement.
Gemini API documentation says the model is generally available and supports a 1M-token context window, a 64K maximum output, and three thinking levels: low, medium, and high. For a long-context application, treat these as published limits rather than a guarantee of output quality or cost fit for your workload. See Google’s current model documentation.
Introductory pricing is listed as $0.75 per 1M input tokens and $3.75 per 1M output tokens through December 31, 2026. The documentation lists higher standard pricing from January 1, 2027. Any cost estimate should therefore state that it uses the introductory window rather than an ongoing price. See the pricing and date details.
Anthropic: what was actually checked
In the official material reviewed through August 23, 2026, Claude Opus 5, announced on July 24, remained the latest displayed general-purpose model announcement available in the supplied research. That specific source does not establish that all of Anthropic’s August activity concerned safeguards, so this article narrows the statement to the material checked rather than generalizing across the company’s full activity. See Anthropic’s official Claude Opus 5 announcement.
The Claude Opus 5 announcement lists a price of $5 per 1M input tokens and $25 per 1M output tokens on Anthropic’s platform. When comparing different providers, keep the same units and separate input pricing from output pricing. Source for the model announcement and pricing.
Shutdown dates: separate past from future
Google’s release notes scheduled Imagen 4 endpoint shutdowns for August 17, 2026; as of August 23, that date has passed. The same notes announce the shutdown of Gemini Robotics ER 1.6 Preview on August 31, 2026, which is still a future date at the time of this review. See the Gemini API release notes.
Operationally, immediately confirm whether any integration was affected by the Imagen 4 shutdown, then assign an owner and migration path for the August 31 Robotics ER shutdown. This is a practical risk-control step, not a claim that every integration will be affected.
How to compare before deciding
Vendor-reported benchmark scores are directional and are not a substitute for a controlled comparison using the same prompts, tools, success criteria, and cost method. This is the editorial methodology for this article, not a claim that the linked OpenAI page independently validates every part of that general rule. See the published OpenAI benchmark context.
- Record the exact model ID and version for every deployment; do not keep only the family name.
- Check that the ChatGPT version is not necessarily the same version used in Codex or ChatGPT Work.
- Run a small set of Arabic and English prompts with measurable success criteria.
- Retest tool calling, including missing or invalid arguments.
- Calculate input and output costs separately, and record whether the price is introductory and when it ends.
- Check deprecation notices before fixing a production default.
SME implementation checklist
- Pin the exact version in each service configuration.
- Keep a record of the check date, price, and lifecycle status.
- Run this article’s editorial Arabic/English test plan before migration.
- Review tool permissions and call behavior after each change.
The appropriate decision is conditional on test results, cost data, and lifecycle status—not a universal endorsement of one model.