Six Labs Hit the Frontier in Eight Days. Your Model Routes Are Already Stale.

Via Artificial Analysis: Four frontier launches in eight days: six labs now field a model above 50 on the Artificial Analysis Intelligence Index

The frontier stopped being a two-horse race

On July 17, 2026, Artificial Analysis reported what every operator with a fixed model route should treat as a wake-up call: four frontier launches in eight days - Grok 4.5, GPT-5.6 (Sol/Terra/Luna), Muse Spark 1.1, and Kimi K3. Six labs now field a model above 50 on the Artificial Analysis Intelligence Index, up from two in early June.

Claude Fable 5 still sits at #1 (about 60), but its lead narrowed from four points to one. The top three now come from three different labs and span just three points. Four of the ten highest-scoring models launched since July 8.

For an owner-led firm, the business consequence is simple: if your production routes still assume last month's cost curve, you are probably overpaying for the same class of work.

Cost per task moved faster than brand loyalty

Artificial Analysis's useful number is not the leaderboard selfie - it is cost per Intelligence Index task. GPT-5.6 Sol (max) lands one point below Claude Fable 5 at roughly $1.04 per task versus about $2.75. Grok 4.5 (high) delivers a 54 at about $0.31. At the 51 band, GPT-5.6 Luna (~$0.21) and Muse Spark 1.1 (~$0.26) undercut what GLM-5.2 priced at that intelligence level a week earlier.

Kimi K3 entered at 57 - third overall - with agentic and knowledge-work scores behind only the top two on several AA suites, at about $0.94 per Index task on the pricing Artificial Analysis cites. Comparable intelligence to Claude Opus 4.8 (max) at roughly half the cost per task, per their write-up.

Near-frontier intelligence got roughly 2-3x cheaper in eight days. That is not a reason to chase every launch. It is a reason to stop treating 'I standardized on Vendor X in Q1' as a forever decision.

The Coding Agent Index moved too: GPT-5.6 Sol in Codex leads at 80 on Artificial Analysis's coding suite, with Terra and Claude Fable 5 close behind depending on harness. That matters if your firm uses AI for document automation adjacent to code-like structured work - templates, transforms, multi-step tool calls - not only chat Q&A.

None of these figures are a promise that Model A will outperform Model B on your client files. They are evidence the market under the summit got denser and cheaper at the same time. Sticky defaults age in days now, not quarters.

What smart firms do instead of model FOMO

• Inventory your routes: which jobs use which model today, and who approved that?
• Re-benchmark one high-volume job and one high-stakes job against two near-frontier alternatives on your real prompts.
• Score full cost per success, not sticker price - include retries and partner review time.
• Keep a fallback: if a lab restricts, reprices, or degrades a model, your workflow should not freeze.
• Document the decision in one page the owner can read: job, model, cost per success, review rule, revisit date.

You do not need six vendors. You need a routing policy that survives an eight-day stampede.

A practical cadence for a 10-50 person firm: monthly route review for high-volume jobs, immediate review when a primary model reprices or a near-peer undercuts cost-per-success by ~30% on your bake-off. Put the owner or ops lead on the calendar invite. If nobody owns the revisit, the stack drifts.

Also separate 'experiment inbox' from 'production routes.' Staff can try new models on non-client sandboxes. Client work stays on documented routes until the bake-off clears the quality bar.

How AgentsROI helps

Model Selection & Continuity Planning is the judgment layer: right model, right job, right privacy/cost tradeoff - with a fallback so a discontinued or repriced model does not break the firm.

If you also cannot prove which workflows deserve frontier spend at all, start with a Workflow ROI Audit. Match capability to jobs that pay; demote everything else to cheaper routes.

Re-check the map, then move

Six labs above 50 is not a trophy for Silicon Valley. It is a reminder that model lock-in decays on a weekly clock. Owners who re-benchmark quietly keep margin. Owners who wait for the next 'AI strategy offsite' keep paying June prices for July intelligence.

If your routes have not been reviewed since early July, ask for a free assessment: Request your free AI assessment.

This article summarizes publicly reported information and is for general informational purposes only. It does not constitute legal, tax, financial, investment, security, or compliance advice. AgentsROI.ai is not a law firm, accounting firm, or registered investment adviser. Facts, pricing, statistics, and product capabilities cited here reflect the sources listed at the time of writing and may change. Readers should verify current information independently and consult qualified professionals regarding obligations specific to their industry, jurisdiction, and circumstances-including applicable New York State and New York City requirements. AgentsROI.ai may have commercial relationships with vendors mentioned; where material, such relationships are disclosed. Nothing in this article is an endorsement of any specific AI product, model, or provider.