#38: Frontier and Open-Weight Model Landscape Update, August 2026

Claude Opus 5 launched on July 24, 2026. Pricing for Opus 5 remained unchanged. Performance results were reported to be strong for agentic coding and computer-use tasks.

Claude Opus 5 was subsequently reported to lead agentic and frontier benchmarks in secondary roundups.

Grok 4.6 was listed with frontier-model pricing and a large context window in August 2026 roundups.

Gemini 3.7 Flash was listed as an efficient-tier model with introductory pricing in August 2026 roundups.

GLM-5.3 was reported as a coding and agent flagship. Its API pricing was reported to be unchanged from the prior generation.

GLM-5.3-Flash was reported with an MIT license. It was described as a 320B MoE model with 18B active parameters. It was reported to have a 1M context window and low pricing.

Qwen3.8-Flash-Next was described as an open-weight Qwen4 preview. It included multimodal support and low pricing.

Tencent Hy4 was reported as a 770B open-sourced flagship. It was noted to have 49B active parameters and a 1M context window.

Hy4 hosted access was reported at a very low per-token rate given its model size.

The summer-2026 open-model survey indicated that Chinese labs consistently shipped the largest open models of the year.

A broad AI comparison page summarized model strengths, context limits, and API pricing across major chatbots.

Grok is described as having fast response times in developer-focused comparison coverage.

Claude, ChatGPT, Gemini, and Grok were compared based on plan pricing and feature bundles in a consumer-facing pricing roundup.

GPT-family, Claude, Gemini, and Grok were presented in a broad model-price comparison that included multiple paid tiers.

No gathered source provides strong primary verification for a new Gemini, Claude, ChatGPT, or Grok release this week.

claim	label	order (🟢 robust / ⚠️ sensitive)	numbers	URL
Claude Opus 5 launched on July 24, 2026 with unchanged pricing and strong agentic coding/computer-use results.	🟢 confirmed	🟢 robust	$5/$25 per 1M tokens; OSWorld 2.0 70.57%	https://www.marktechpost.com/2026/07/24/meet-the-new-claude-opus-5-frontier-class-agentic-coding-and-computer-use-at-unchanged-opus-pricing/
Claude Opus 5 was reported to lead agentic and frontier benchmarks in secondary roundups.	🟡 partial	⚠️ sensitive	43.3% Frontier-Bench v0.1; Agentic Index first	https://betterstack.com/community/guides/ai/claude-opus-5/
Grok 4.6 was listed with frontier-model pricing and a large context window in August 2026 roundups.	🟡 partial	⚠️ sensitive	$2/$6 per 1M tokens; 500K context	https://capitalandcompute.net/blog/new-ai-models-august-2026/
Gemini 3.7 Flash was listed as an efficient-tier model with introductory pricing in August 2026 roundups.	🟡 partial	⚠️ sensitive	$0.75/$3.75 per 1M tokens	https://capitalandcompute.net/blog/new-ai-models-august-2026/
GLM-5.3 was reported as a coding and agent flagship with unchanged API pricing from the prior generation.	🟡 partial	⚠️ sensitive	$1.40/$4.40 per 1M tokens	https://themodelgap.com/new-ai-models/2026-08
GLM-5.3-Flash was reported as MIT-licensed, 320B MoE, 18B active, with 1M context and low pricing.	🟡 partial	⚠️ sensitive	320B total; 18B active; $0.15/$0.50 per 1M	https://themodelgap.com/new-ai-models/2026-08
Qwen3.8-Flash-Next was described as an open-weight Qwen4 preview with multimodal support and low pricing.	🟡 partial	⚠️ sensitive	125B total; 6B active; $0.15/$0.47 per 1M	https://capitalandcompute.net/blog/new-ai-models-august-2026/
Tencent Hy4 was reported as a 770B open-sourced flagship with 49B active parameters and 1M context.	🟡 partial	⚠️ sensitive	770B total; 49B active; 1M context	https://www.forbes.com/sites/jonmarkman/2026/08/31/tencent-open-sources-hy4-its-770-billion-parameter-flagship-model/
Hy4 hosted access was reported at a very low per-token rate for a model of its size.	🟡 partial	⚠️ sensitive	$0.83 per 1M input tokens	https://www.forbes.com/sites/jonmarkman/2026/08/31/tencent-open-sources-hy4-its-770-billion-parameter-flagship-model/
The summer-2026 open-model survey said Chinese labs repeatedly shipped the largest open models of the year.	🟡 partial	🟢 robust	754B to 2.78T monthly ceilings	https://huggingface.co/blog/state-of-open-models-summer-2026
The broad AI comparison page summarized model strengths, context limits, and API pricing across major chatbots.	🟡 partial	⚠️ sensitive	no single verified total	https://penchan.co/en/ai/compare/
Grok is described as having fast response times in developer-focused comparison coverage.	🟡 partial	⚠️ sensitive	1–4 s; <1 s in one table	https://techlifeadventures.com/post/grok-vs-claude-vs-chatgpt-vs-gemini-developers-2026
Claude, ChatGPT, Gemini, and Grok are compared on plan pricing and feature bundles in a consumer-facing pricing roundup.	🟡 partial	⚠️ sensitive	$20/$20/$19.99/$30 mo	https://aitoolsreview.co.uk/insights/claude-vs-chatgpt-gemini-grok
GPT-family, Claude, Gemini, and Grok are presented in a broad model-price comparison with multiple paid tiers.	🟡 partial	⚠️ sensitive	$8–$249.99/mo tiers	https://rg2b.com/chatgpt-vs-claude-vs-gemini-vs-grok-2026/
No gathered source gives strong primary verification for a new Gemini, Claude, ChatGPT, or Grok release this week.

To do:

Act: Evaluate Claude Opus 5 for agentic workflows, particularly in coding and computer automation. The reported performance on OSWorld 2.0 suggests this may be a strong candidate for such uses.
Watch: Monitor the performance and adoption of Tencent Hy4, especially given its reported open-source nature and large parameter count combined with low hosted access pricing. Its large context window also warrants attention.
Ignore: Claims of new releases from major frontier model providers (Gemini, Claude, ChatGPT, Grok) this week lack strong primary verification. Focus on confirmed launches and model updates.