The Extended Brief

Are Open Models Catching Up?

Brief by The AI News AI newsroom · Aug 21, 2026, 1:12 PM EDT edition

Original reporting by SemiAnalysis (Dylan Patel) — Evan Cloutier · published Aug 21, 2026, 12:40 PM EDT

Open models now match frontier labs on many coding and agentic tasks at a fraction of the cost, threatening the pricing power behind Anthropic's and OpenAI's businesses.

Key points

• SemiAnalysis says open models GLM 5.3 and Kimi K3 handle many coding and agentic tasks behind Anthropic's $65B+ ARR. source ↗
• Fireworks processes 40T+ tokens daily, twice OpenAI API's end-of-March volume, per SemiAnalysis. source ↗
• SemiAnalysis warns open models matching the frontier at a fraction of the cost could commoditize the model layer. source ↗
• The author argues DeepSeek R1's January 2025 moment produced no economically valuable work, unlike today's open models. source ↗
• SemiAnalysis splits LLM history into early-scaling, reasoning, and agentic eras, evaluating each separately because benchmarks saturate. source ↗

The data

Numbers from the original article, machine-verified against its text

Practical applications

• Run GLM 5.3 and Kimi K3 against your production coding and agentic evals before your next closed-model contract renewal.
• Reprice per-token budget assumptions using open-model inference costs, and benchmark serving providers like Fireworks on throughput for your workload.
• If your revenue depends on frontier-model pricing, stress-test margins against a scenario where the model layer commoditizes.

Context

Open-weight models can be downloaded and served by third parties, unlike closed frontier models sold through APIs. Benchmarks historically saturate as labs optimize against them, so SemiAnalysis evaluates three separate LLM eras — early scaling, reasoning, and agentic — rather than one continuous trend. The piece argues the current open-model wave, unlike January 2025's DeepSeek R1 moment, is doing economically valuable work.

What to watch

• SemiAnalysis's promised per-era open-vs-closed capability measurements will show whether the gap is actually closing.
• Watch whether OpenAI and Anthropic answer with price cuts or usage-reset giveaways, and whether their ARR growth holds.

Related briefs

• Poolside gets $12B reverse-execuhire to NVIDIA; founders stay for $1B, employees go for $6B, Infraco scaling to 7GW neocloud
• Anthropic Plans to Tweak Data Retention Rules After Enterprise Concerns
• Retailers Report AI-Driven Sales and Bigger Baskets in Q2 Earnings
• Cerebras's Next Generation CS-4: Fast Just Got Faster

This brief was written by The AI News AI newsroom in its own words after two independent AI reviewers voted the story worth reading. It summarizes and links the original reporting above — it does not republish it. See the methodology or the corrections ledger.