AI News HubLIVE
In-site rewrite2 min read

Show HN: Tracking GenAI cost and endpoint fragility so app teams don't have to

LLMIntel is a demo dashboard for monitoring GenAI model usage costs, endpoint health, and optimization opportunities. It provides views for model status, cost analysis, usage trends, at-risk spend, and tag breakdowns, helping teams take action before model deprecation or cost spikes.

SourceHacker News AIAuthor: ATsimbalistov

A sample LLMIntel dashboard with a week of realistic traffic across three apps.

You’re viewing demo data — this is what LLMIntel looks like with a week of real traffic. Sign up and your own data appears minutes after your first instrumented call.

Get your API key →2-minute quickstart

Overview

6

Models you run

1

Need attention (≤90d / deprecated)

$0.05–5.32

Input $/1M spread

$0.4–16

Output $/1M spread

Lifecycle health

ModelStateRetirement

meta.llama3-1-405b-instruct-v1:0retiringin 21 days2026-07-28

Cost analysis

Pricing known for 6 of 6 models you run. Cheapest by input: gpt-5-nano-2025-08-07 at $0.05/1M.

Want cheaper/faster on-par alternatives for these models? See optimization →

Cost & usage

Times in UTC · 13 months of history on your plan · sample data — send telemetry to see your own, range-filtered

Filters

ProviderModelTag

Spend trend

Stack by

$1,284.06

Tracked spend · last 30 days ▲ +9.9%

412,000

Requests ▲ +7.5%

96M

Total tokens ▲ +11%

$13.32

Blended rate · $/1M tokens

−$48.74

Caching savings (est.)· 27M cached

$212.40

At-risk spend · retiring ≤90d

Cost is frozen at ingest — priced from each model's rates at call time, so past spend never shifts. Window: last 30 days.

Telemetry up to Jul 7, 16:41 UTC· 8 min ago· Catalog checked 42 min ago

Responses getting longer

These models are returning more output per input token than their 30-day baseline — the usual cause of silent cost creep. Check prompts, max_tokens, and whether a cheaper or more concise model would do.

ModelNowBaselineDriftRequestsSpend

claude-sonnet-4-5-202509290.3×0.2×+38%101,040$620.86

At-risk spend

Dollars flowing through models that retire within 90 days. Migrate before the provider 4xxs you — see recommended replacements on each model page.

ModelRetirementSpend

meta.llama3-1-405b-instruct-v1:0in 21 days2026-07-28$212.40

Spend by source

Environment › Application — expand any row to drill in

Group byFilter tag

Environment › ApplicationRequestsTokensBlended $/1MCost/reqSpendShare

prod379,00086M$13.16$0.0030$1,131.6688%

Checkout Assistant261,00058M$12.77$0.0028$742.1158%

Support Copilot118,00028M$13.96$0.0033$389.5530%

staging33,00010M$14.65$0.0046$152.4012%

Unassignedno app mapping33,00010M$14.65$0.0046$152.4012%

How is this allocated?

Choose the root dimension with Group by; the tree nests it over the next dimension in the rotation Environment → Application → API key. An application comes from the app tag your agent sends or the API key's app mapping; calls with neither show as Unassigned (map keys under Applications or send an app tag). Unattributed keys are usage recorded before per-key attribution shipped. Nothing is ever smeared across buckets, and the tree follows the active filters (including the tag filter above). Share is relative to the total spend in view.

Custom tag breakdowns

grouped by tag key

Tag: feature4 values

featureRequestsTokensBlended $/1MCost/reqSpendShare

checkout214,00048M$12.51$0.0028$604.3247%

search121,00028M$14.54$0.0033$401.1831%

summarize54,00013M$14.68$0.0035$187.9615%

Untaggedno feature tag23,0007.7M$11.77$0.0039$90.607%

Tag: team3 values

teamRequestsTokensBlended $/1MCost/reqSpendShare

growth268,00061M$12.75$0.0029$771.2460%

platform144,00033M$13.70$0.0031$452.2235%

Untaggedno team tag04.3M$14.09—$60.605%

Spend by model

ModelProviderRequestsInputOutputBlended $/1MCost/reqTokens/reqOut:inSpend

claude-sonnet-4-5-20250929Anthropic168,40031M

14M cached

8.5M$15.64$0.00372360.3×$620.86

gpt-5-2025-08-07OpenAI96,80022M

9.0M cached

6.9M$9.98$0.00303030.3×$292.40

gpt-5-mini-2025-08-07OpenAI74,30012M

3.6M cached

3.4M$2.09$0.00042090.3×$32.40

gpt-5-nano-2025-08-07OpenAI41,9005.6M1.5M$1.13$0.00021690.3×$8.00

gpt-4o (2024-08-06)Azure AI Foundry22,6002.9M900k$31.05$0.00521680.3×$118.00

meta.llama3-1-405b-instruct-v1:0AWS Bedrock8,000780k220k$212.40$0.02661250.3×$212.40

Optimization

Optimization

Analysis: Artificial Analysis · advisory

gpt-5-2025-08-07

GPT-5 mini

input −80% · output −80% · same capability class

Compare

meta.llama3-1-405b-instruct-v1:0

Claude Sonnet 4.5switch to anthropic

current model retiring ≤90d · higher capability class · context 2.5×

Compare