The Markets Desk — Week 31 · Tue 28 Jul 2026

Markets

NVDA 196.51 −4.99%·AVGO 383.22 +0.34%·MU 900.20 −2.25%·AMAT 516.89 −3.61%·LNZA 5.51 −0.36%·ASTS 58.29 +3.72%·FN 470.82 −1.10%·MDT 84.22 +1.21%·TSHA 5.54 +0.36%·USAR 14.93 +5.51%·WULF 17.98 −2.60%·MNTS 4.79 +1.70%·LTBR 8.02 +6.93%·A 137.89 −0.49%·ADI 371.89 +0.01%·CRCL 65.67 +5.31%·DFDV 2.73 +1.11%·MAZE 28.69 +3.65%·PONY 7.38 +3.51%·BIDU 105.29 −0.05%·LUNR 13.31 +3.02%·CRWV 70.79 −1.52%·OKLO 41.81 +3.88%·NAKA 4.69 −3.70%·SNPS 388.99 +4.16%·XPEV 12.78 +3.65%·ARAY 0.23 −5.62%·CSCO 114.57 +0.35%·KEYS 313.68 −1.57%·MGX 1.21 +0.41%·NVDA 196.51 −4.99%·AVGO 383.22 +0.34%·MU 900.20 −2.25%·AMAT 516.89 −3.61%·LNZA 5.51 −0.36%·ASTS 58.29 +3.72%·FN 470.82 −1.10%·MDT 84.22 +1.21%·TSHA 5.54 +0.36%·USAR 14.93 +5.51%·WULF 17.98 −2.60%·MNTS 4.79 +1.70%·LTBR 8.02 +6.93%·A 137.89 −0.49%·ADI 371.89 +0.01%·CRCL 65.67 +5.31%·DFDV 2.73 +1.11%·MAZE 28.69 +3.65%·PONY 7.38 +3.51%·BIDU 105.29 −0.05%·LUNR 13.31 +3.02%·CRWV 70.79 −1.52%·OKLO 41.81 +3.88%·NAKA 4.69 −3.70%·SNPS 388.99 +4.16%·XPEV 12.78 +3.65%·ARAY 0.23 −5.62%·CSCO 114.57 +0.35%·KEYS 313.68 −1.57%·MGX 1.21 +0.41%·

The Record14 callsFull track record →

Token Price Index

PRIMARY-SOURCED FROM PROVIDER PRICING PAGES · OBSERVED 20 JUL 2026

If you’re shipping AI products on API inference, you’re paying token prices you can’t independently check — the invoice is usually the first place a team learns a model repriced. This is the hand-verified price of intelligence: every major model normalized to blended $/Mtok by capability tier, so you can see whether you’re overpaying before the invoice does. The hero metric is the open-vs-closed spread — how much cheaper you run open weights at the same capability bar.

6.21×

cheaper to run open weights at the Strong tier, blended. The frontier has no open peer — that’s the premium the whole economy hangs on.

Stack Watch

Free — the alert tier of Managed Inference

Get the alert when the rungs move

Every new model lands on the index the day its price verifies — GPT-5.6’s three rungs landed July 16. We re-sweep on every move and email you if the cheapest option at any capability tier changes. No account, no spam.

Open / Closed Spread by Tier

Blended $/Mtok · 3:1 input:output profile

FrontierBEST AVAILABLECLOSED $4.50NO OPEN PEER—NO SPREAD

StrongNEAR-FRONTIERCLOSED $3.38OPEN $0.5446.21×OPEN CHEAPER

MidWORKHORSECLOSED $0.563OPEN $0.1753.21×OPEN CHEAPER

SmallCHEAP & FASTCLOSED $0.175OPEN $0.062.92×OPEN CHEAPER

All Tracked Models

$/Mtok · read straight from pricing pages

Model	Provider	Tier	Weights	Input	Output	Blended
Gemini 3.1 Pro	google	FRONTIER	CLOSED	2.00	12.00	4.50
GPT-5.6 Sol	openai	FRONTIER	CLOSED	5.00	30.00	11.25
Claude Opus 4.8	anthropic	FRONTIER	CLOSED	5.00	25.00	10.00
GPT-5.5	openai	FRONTIER	CLOSED	5.00	30.00	11.25
Claude Fable 5	anthropic	FRONTIER	CLOSED	10.00	50.00	20.00
DeepSeek V4 Pro	deepseek	STRONG	OPEN	0.43	0.87	0.54
Gemini 3.5 Flash	google	STRONG	CLOSED	1.50	9.00	3.38
Claude Sonnet 5	anthropic	STRONG	CLOSED	2.00	10.00	4.00
GPT-5.6 Terra	openai	STRONG	CLOSED	2.50	15.00	5.63
GPT-5.4	openai	STRONG	CLOSED	2.50	15.00	5.63
Llama 3.1 405B	meta	STRONG	OPEN	3.50	3.50	3.50
DeepSeek V4 Flash	deepseek	MID	OPEN	0.14	0.28	0.18
Gemini 3.1 Flash-Lite	google	MID	CLOSED	0.25	1.50	0.56
Llama 4 Maverick	meta	MID	OPEN	0.27	0.85	0.42
Gemini 2.5 Flash	google	MID	CLOSED	0.30	2.50	0.85
GPT-5.4-mini	openai	MID	CLOSED	0.75	4.50	1.69
Llama 3.3 70B	meta	MID	OPEN	0.88	0.88	0.88
GPT-5.6 Luna	openai	MID	CLOSED	1.00	6.00	2.25
Llama 3.2 3B	meta	SMALL	OPEN	0.06	0.06	0.06
Llama 4 Scout	meta	SMALL	OPEN	0.08	0.30	0.14
Gemini 2.5 Flash-Lite	google	SMALL	CLOSED	0.10	0.40	0.18
GPT-5.4-nano	openai	SMALL	CLOSED	0.20	1.25	0.46
Claude Haiku 4.5	anthropic	SMALL	CLOSED	1.00	5.00	2.00

Estimate Your Bill

10M input / 2M output per month, Strong tier

closed $33.00/moopen (hosted) $6.09/moyou save 82%

Capability tier needed

Monthly tokens (millions)

Workload shape

Cached input: 0%

Cheapest closed$337.50/moGemini 3.5 Flash

Cheapest open$54.38/moDeepSeek V4 Pro

$283.13/mo

saved by running open — 84% cheaper

We do this for real — token-cost audit →

Or Have It Handled

Managed Inference — the standing version of the auditSee Managed Inference →

£500/mo

Your bill won’t stay optimized — providers repriced or changed terms 32 times last quarter. Managed Inference re-benchmarks your actual model mix against this index every month and sends the switch list, with an alert any week your number moves. Built for teams spending £5k+/month on inference.

Price Is Half the Picture

The rest of the sovereignty desk

Terms

Provider Terms Index

Who trains on your data, who can change price mid-contract — every major provider scored against its own terms.

UPDATED JUL 2026 Lock-in

Lock-in Score

How locked in are you? A five-minute audit of your switching cost, scored 0–100.

TAKE THE SCORE →Managed

Managed Inference

Prefer it handled? We run the stack against the index — £500/mo retainer.

STACK WATCH FREE

Methodology

Every price is read from the provider’s own public pricing page and dated — never a reseller. Blended $/Mtok weights input:output by workload profile; tiers are assigned by benchmark band and re-checked at each model release. One trap folded in: newer tokenizers emit more tokens for the same text, so compare price per task, not per raw token. Last observed 20 JUL 2026, re-swept monthly.