# Frontier AI models: August 2026

> No new flagship - the story moved to inference economics, and OpenAI's custom Jalapeño chip put a number on how much cheaper automation is about to get.

## Direct answer

No frontier model launched in August 2026. The current frontier models are Claude Opus 5 (Anthropic), GPT-5.6 Sol (OpenAI) and Gemini 3.6 Flash (Google), all released in July. August's significant development was economic: OpenAI's disclosed losses of roughly $33B against a $25B revenue run rate, and its Broadcom-built Jalapeño inference chip claiming about 50% lower cost per token. Cost per token, not capability, is now the variable that decides which jobs get automated next.

## Released in August 2026

| Date | Model | Lab | Capability | Effect on automation |
| --- | --- | --- | --- | --- |
| 4 August 2026 | Jalapeño inference chip | OpenAI and Broadcom | Custom inference silicon, claimed ~50% lower cost per token | Halving inference cost halves the price of every automated task, which reaches far further down the job ladder than any single model release. |

## Leading models as of August 2026

| Lab | Leading model | Strongest at |
| --- | --- | --- |
| Anthropic | Claude Opus 5 | Frontier intelligence at half the previous price |
| OpenAI | GPT-5.6 Sol | Autonomous long-horizon agent work |
| Google | Gemini 3.6 Flash | Efficient high-volume agentic tasks |

## What this month changed for careers

What protects most jobs right now is not skill - it is price. AI is already capable of far more than employers currently pay it to do. August 2026 made clear that the price side of that equation is falling fastest, which means the safest-feeling roles can change status without any new model being released at all.

## Questions

### What are the current frontier AI models in August 2026?

Claude Opus 5 from Anthropic, GPT-5.6 Sol from OpenAI and Gemini 3.6 Flash from Google. All three were released in July 2026 and no new flagship shipped in August.

### What is OpenAI's Jalapeño chip?

Jalapeño is OpenAI's custom inference chip, built with Broadcom and disclosed in August 2026. It is claimed to deliver roughly 50% lower cost per token than the GPUs it replaces, which directly reduces the cost of running any automated task.

### Why does inference cost matter more than model capability for jobs?

Because capability sets what is possible and price sets what actually happens. Most tasks that AI can technically perform are not automated yet, because the business case does not clear. Falling inference cost clears those cases without any new capability being added.

---

**Source:** Job Security Meter — https://jobsecuritymeter.com/ai-model-releases/august-2026
**Human-readable version:** https://jobsecuritymeter.com/ai-model-releases/august-2026
**Last updated:** August 2026
**Attribution:** Free to quote and cite with attribution to Job Security Meter.
**Full site specification for language models:** https://jobsecuritymeter.com/llms-full.txt