“Cheapest” and “best value” are different questions. Value means capability per dollar — and August 2026’s list is unusually interesting because of OpenAI’s price cuts and Anthropic’s introductory rate.
The August Value Picks
| Model | Price $/M in/out | Why It’s a Value Pick |
|---|---|---|
| GPT-5.6 Luna | $0.20/$1.20 | Flagship-class coding at budget prices |
| Claude Sonnet 5 (intro) | $2/$4 | Top mid-tier agentic model until Aug 31 |
| Grok 4.5 | $2/$6 | Cheapest production-grade (score 70+) |
| Kimi K3 | $3/$15 | Cheapest frontier-tier (score 80+) |
| DeepSeek V4 Pro | $0.435/$0.87 | Strongest open-weight reasoning per dollar |
Why These Win
Luna and Sonnet 5 benefit from temporary/structural pricing moves this month. Kimi K3 and Grok 4.5 anchor their tiers at the exact price-performance frontier. DeepSeek V4 Pro remains the open-weight ratio play with MIT licensing.
The Framework
Score your actual tasks, not vendor benchmarks: measure tokens per accepted result, retry rates, and latency for your specific workload. A mid-tier model that needs 2x retries loses to a pricier one that succeeds first try.
Bottom Line
If you buy this month: lock in Sonnet 5’s intro rate before August 31, re-evaluate OpenAI at the new Luna/Terra rates, and use Kimi K3 or Grok 4.5 as the value anchors of any multi-provider strategy. Full methodology in our AI Model Comparison 2026.
