A Survey of the Frontier

The frontier is a moving line, not a badge. Herein its position as of August 2026 — sixteen models in service, passage costs per 1M tokens, every figure dated and cross-checked.

VOL. 1 · BUILDER EDITION · RE-SURVEYED EACH RELEASE CYCLE

FLAGSHIP RANGE

PRODUCTION PLAINS

OPEN COUNTRY

THE FRONTIER LINE — AUG 2026

Claude Opus 5 ⚑

high ground · idx 61 · $25

LEGEND

independent mark (Artificial Analysis)

current lead, flag planted

the frontier line, surveyed Aug 2026

EAST: OUTPUT $/1M, LOG ($1·$3·$10·$30)
ELEVATION: AA INTELLIGENCE INDEX

~1/wk

new frontier releases charted, summer 2026

parties camped in the top Arena-Elo tier, Mar 2026

~110×

spread in passage cost, cheapest to priciest token

8 pts

top closed over top open on the AA index — and closing

−80%

steepest single price cut of the year — Luna, Jul 30

Register of Active Expeditions

16 MODELS IN SERVICE · PRICES PER 1M TOKENS, IN / OUT · VENDOR FIGURES FLAGGED

Flagshipclosed · deepest reasoning

01 Claude Opus 5 Anthropic Jul 24, 2026 ~1M $5 / $25 61 closed indep. Thinking on by default; adds a Fast mode. Half the input price of Fable 5. 02 Claude Fable 5 Anthropic Jun 9, 2026 ~1M ~$10 / — 60 closed indep. Access suspended Jun 12–30 under a US export-control directive, then fully restored Jul 1. 03 GPT-5.6 Sol OpenAI GA Jul 2026 1.05M $5 / $30 59 closed indep. Long-context surcharge above 272K input tokens. Fast mode available. 04 Gemini 3.1 Pro Google Feb 2026 ~1M $2 / $12 — closed indep. The newer Gemini 3.5 Pro (rumored 2M context) remains unreleased as of August. 05 Grok 4.6 xAI Aug 12, 2026 ~500K above 4.5 rates — closed indep. Priced above Grok 4.5; independent numbers still landing. Context surcharge on long inputs.

Balancednear-flagship · production economics

06 GPT-5.6 Terra OpenAI GA Jul 2026 1.05M $2 / $12 — closed indep. Cut 20% on Jul 30. Production default for many teams. 07 Claude Sonnet 5 Anthropic Jun 30, 2026 ~1M ~$2 (intro) — closed indep. The default workhorse for many Claude builders. 08 Gemini 3.7 Flash Google Aug 13, 2026 1M ~$0.75 / ~$3.75 (intro) — closed vendor Intro pricing through end of 2026, then doubles Jan 1, 2027. Headline benchmarks vendor-reported. 09 Grok 4.5 xAI Jul 8, 2026 ~500K $2 / $6 54 closed indep. Context surcharge above 200K roughly doubles rates.

Efficientfast · cheap · high-volume

10 GPT-5.6 Luna OpenAI GA Jul 2026 1.05M $0.20 / $1.20 — closed indep. Cut 80% on Jul 30 — the steepest single cut of the year from a US lab. 11 Claude Haiku 4.5 Anthropic 2025 ~200K low-cost tier — closed indep. Optimized for speed and cost rather than deepest reasoning.

Open-weightdownload · self-host · fine-tune

12 DeepSeek V4-Pro DeepSeek Apr 24, 2026 1M $0.44 / $0.87 53 MIT indep. Text-focused; peak-hours surcharge. Powers the free DeepSeek app. 13 DeepSeek V4-Flash DeepSeek 2026 1M $0.14 / $0.28 — MIT indep. Lighter 284B/13B sibling of V4-Pro; near it on many coding tasks. 14 Kimi K3 Moonshot AI Jul 17, 2026 ~1M ~$0.6 / ~$3 — Modified MIT vendor Adds native vision. Hosted price varies by provider. 15 Qwen3.8-Max Alibaba Aug 3, 2026 1M $2 / $6 — open vendor Priced well above DeepSeek; GA rate card supersedes July preview discount. 16 GLM-5.3 Z.ai Aug 14, 2026 1M GLM Coding Plan / TBD — open vendor Launched Aug 14; open weights/API staged behind safety review. Benchmarks thin, vendor-only.

THE FRONTIER MODEL REPORT — VOL. 1 FRONTIERMODEL.IO