The Markets Desk — Week 31 · Tue 28 Jul 2026

Markets

NVDA 193.64 −1.46%·AVGO 374.41 −2.30%·MU 792.21 −12.00%·AMAT 468.78 −9.31%·LNZA 5.56 +0.91%·ASTS 53.69 −7.89%·FN 446.33 −5.20%·MDT 88.92 +5.58%·TSHA 5.40 −2.53%·USAR 13.54 −9.31%·WULF 16.27 −9.51%·MNTS 4.53 −5.43%·LTBR 7.82 −2.49%·A 142.32 +3.21%·ADI 366.15 −1.54%·CRCL 61.90 −5.74%·DFDV 2.53 −7.33%·MAZE 28.26 −1.50%·PONY 7.32 −0.81%·BIDU 103.47 −1.73%·LUNR 12.01 −9.77%·CRWV 65.16 −7.95%·OKLO 38.04 −9.02%·NAKA 4.39 −6.40%·SNPS 387.01 −0.51%·XPEV 12.75 −0.23%·ARAY 0.23 +4.00%·CSCO 113.73 −0.73%·KEYS 298.60 −4.81%·MGX 1.20 −0.71%·

The Record14 callsFull track record →

Your model mix is priced against a market that changed under you 32 times last quarter — new prices, deprecations, terms shifts — and you usually find out when the invoice lands. A one-off optimization pass has a shelf life; the target keeps moving after you've stopped looking. Managed Inference is the standing version of the audit: once a month we re-benchmark your actual model mix against current prices and send the switch list, plus an alert any week your number moves past your threshold.

Monthly re-benchmark memo

Your stack recomputed against the latest index: what moved, what to switch, the saving and the risk you dodged, the reasoning.

Threshold alerts

The same engine as Stack Watch, but on your actual model mix — any week the number moves past your threshold.

Async questions

Email us, fair use, answered within 2 working days. No calls, no standing meetings.

What a switch list looks like

Illustrative — live Index rows, not a client outcome

Strong tier · Claude Sonnet 5 — $2.00 / $10.00 per Mtok today, but intro pricing ends 31 Aug and it steps up to $3.00 / $15.00 on 1 Sep. A scheduled 50% rise on a rung you're already standing on — flagged before it lands, not after.
Same capability bar, open weights — DeepSeek V4 Pro runs $0.435 / $0.87 per Mtok. Whether your workload can tolerate that spread is the judgment call the memo makes for you.
Workhorse tier — GPT-5.4-mini ($0.75 / $4.50) versus Gemini 3.1 Flash-Lite ($0.25 / $1.50): same tier, 3× apart on input.

Prices observed 6–16 Jul 2026, read by hand from each provider's own pricing page — the independent index nobody else maintains. Your memo runs the same comparison on your actual model mix.

£500/mo

Built for teams spending £5k+/month on inference.

First 3 founding clients — £350/mo, locked.

First month free — a full re-benchmark and switch list before you pay a penny. Your first paid month is guaranteed too: if it doesn't surface saving or a live risk worth more than the fee, we refund it, no argument. £500/mo, cancel anytime.

Already run the £500 Token-Cost Audit? It credits into your first Managed month if you subscribe within 30 days — you're not paying twice to taste it.

Opening soon — tell us your stack and we'll reach out.