Skip to content
Model Pulse
Reseller moveSnapshot · data through Sep 14, 2026selected by: newest qualifying gateway move over 40%

Eden AI raised its DeepSeek V4 Flash 0731 markup by 487%

The lab has not moved — this is a reseller changing its margin, from $0.03 to $0.176 per million input tokens. Worth knowing if you buy through them.

Compare every provider →Today’s changes →

Independent price and benchmark analysis for AI models — what a workload actually costs, and where to buy it.

The wire

What was written · what actually changed

Full changelog →
In the pressmodel-linked coverage
All news →
  1. Abacus.AI Smaug Models Cut Agent Costs 100x [2026] ↗
    shattered.io· 1d agoQwen3.8 27B
  2. The TechBeat: AI Coding Tip 035 - Split Every Skill Description Into Three Sentences (9/12/2026) ↗
    hackernoon.com· 2d agoQwen3.8 27B
  3. DeepSeek Releases V4.1 Flash, a 748-Billion Flash That Now Sees ↗
    pasqualepillitteri.it· 2d agoQwen3.8 Flash Next
  4. Qwen2.5-Coder-32B-Instruct vs Qwen3.8 Max 0902 - AI Model ... ↗
    opencode.ai· 3d agoQwen2.5-Coder-32B-Instruct
  5. DeepSeek's new model sets a template for powerful LLMs that run lean ↗
    theregister.com· 3d ago
  6. I Built a 256GB “E-Waste” AI Rig to Escape Token Fees. Here’s What Nobody Tells You. ↗
    vocal.media· 4d agoQwen3.8 Flash Next
The labsfirst party · the model itself repriced
  1. 01DeepSeek V4 Flash$0.14 → $0.15 /M · DeepSeek▲ 7%
  2. 02DeepSeek V4 Flash Vision Exp$0.14 → $0.15 /M · DeepSeek▲ 7%
The streetresellers · markup moved, the model did not
  1. 01Solar Pro 4$0.03 → $0.09 /M · OpenRouter▲ 200%
  2. 02Kimi K2.6$0.22 → $0.6 /M · DevPass (LLM Gateway)▲ 173%
  3. 03Qwen3 Coder Next$0.22 → $0.5 /M · Amazon Bedrock▲ 127%
  4. 04DeepSeek V3.2$0.28 → $0.62 /M · Vercel AI Gateway▲ 121%
  5. 05Qwen3-Coder 480B-A35B Instruct$0.22 → $0.45 /M · Amazon Bedrock▲ 105%

Same weights. Wildly different bill.

The gap nobody else publishes · chat workload

The exchange →

55 of 61 providers publish a price for GPT OSS 120B. The cheapest full-context option is LLM Gateway at $0.032/$0.14 per M. Listed prices span 50.3× for identical weights.

Prices use the chat workload: 2K input + 500 output tokens, 8K context. A model is not one price — it is a spread across every venue that sells it. Listings more than 20× from the model’s own median are dropped as feed errors before anything reaches this board, and a cheapest quote counts only when a second venue is within 1.5× of it.

The boards

Four cuts from today's data · prices use the chat workload

Work the numbers yourself

Three questions · each answered with a figure computed today

What will it cost?

Sticker price is a rate, not a bill.

Pick the shape of your traffic — chat, RAG, agentic, long-context — and get the monthly number, cache writes and context tiers included.

$0.306

cheapest workhorse · Qwen3.5 122B-A10B

Price a workload →

Is it any good?

79% of published scores are not independently measured.

We rank on independently measured results only, and price each benchmark for the token profile it actually exercises.

29,633

pts/$ leader · DeepSeek Reasoner on Aider Polyglot

See the evidence →

Should I host it?

The rent is the floor, and it is higher than people expect.

We compute the smallest configuration the weights fit on. Throughput is not knowable from a model name, so every break-even here is a lower bound.

$1,314

lowest break-even · Qwen3.6 27B on 1× A100 80GB

Price out self-hosting →

The ecosystem watch

GitHub activity + npm usage · discovery signals, not recommendations

Full ecosystem →
Mastraframework · 3 signals83/100Ollamainfrastructure · 2 signals76/100Open WebUIapplication · 2 signals76/100browser-useagent · 2 signals75/100

Weekly edition

One email. The changes that actually cost you money.

1159 catalog changes landed in the last 24 days. Most are inventory noise. The digest separates first-party repricing and retirements from reseller noise, names the models whose bill moved, and says plainly when the week was quiet.

Auto-written from the auditable diff log · no manufactured story · unsubscribe in one click

Subscribe via RSS

Email edition coming; RSS carries the same digest today.

Read this week's →RSS →For agents →