A price that doubled and then halved, and an MCP server that flipped six times: TANDIU weekly data notes (Nov 2–9, 2026)
Each week we summarize what changed in the public data on TANDIU: LLM API list prices, AI funding rounds, GPU rental prices and official MCP servers. Every number below comes from the site or its MCP server, as of Novemb
Each week we summarize what changed in the public data on TANDIU: LLM API list prices, AI funding rounds, GPU rental prices and official MCP servers. Every number below comes from the site or its MCP server, as of November 9, 2026 (UTC+8).
1. LLM API list prices
We track 72 models. One official list price changed this week, and it undid last week's change:
| Date | Model | Blended price, USD per 1M tokens | Change |
|---|---|---|---|
| Oct 29 | Mistral Large 4 Preview | $1.03 → $2.07 | +100% |
| Nov 4 | Mistral Large 4 Preview | $2.07 → $1.05 | −49.2% |
Blended price = (3 × input + 1 × output) / 4. For a team that picked this model on price, its cost doubled and then halved inside one week. That is why we check official pricing pages every day and keep every change on record.
Where the range sits today (input / output, USD per 1M tokens):
- Lowest: Hunyuan A13B at $0.075 / $0.30
- Lowest from a US vendor: GPT-6 Luna at $0.10 / $0.50
- Highest: GPT-5.5 Pro and GPT-5.4 Pro at $30 / $180
- Median blended price: $2.00
Gemini 4 Argon is still on its introductory price until November 30, after which $4 / $20 applies.
Source: Token market
2. Price index
The TANDIU LLM price index (base 100 on July 13, 2026; chain-linked, so adding or retiring models does not move it) stands at 105.635, up 0.77% over 30 days. The Nov 4 Mistral cut pulled it back from 106.632.
Source: Price index
3. AI funding
The largest rounds dated this week, all with status "announced":
| Date | Company | Amount | Valuation |
|---|---|---|---|
| Nov 6 | OpenAI | $11B | $1.25T |
| Nov 6 | Thinking Machines Lab | $5B | $40B |
| Nov 5 | Zipline | $1.5B | $21.5B |
| Nov 7 | CesiumAstro | $730M | over $4.5B |
| Nov 7 | Harbinger | $240M | not disclosed |
One note on status: on Nov 8 the Thinking Machines Lab round was in our data as "reported" (talks, unconfirmed). It is now "announced". Every record has a status field, so an agent can count only confirmed rounds without reading the articles.
Source: Funding
4. GPU cloud rental
Median rental price on the vast.ai public market (international, not a China purchase price), USD per GPU-hour, with change versus 7 days earlier:
| GPU | Median today | vs 7 days ago |
|---|---|---|
| B200 180GB | $8.75 | +40% |
| H200 141GB | $4.34 | +32% |
| H100 SXM5 80GB | $2.67 | +11% |
| RTX PRO 6000 Blackwell 96GB | $1.44 | −2% |
| L40S 48GB | $0.54 | −33% |
| RTX 5090 32GB | $0.50 | −14% |
Data-center Hopper and Blackwell cards got more expensive to rent this week, while workstation and consumer cards got cheaper. Sample sizes are small for the top cards (B200: 9 listings), so treat the moves as indicative.
Source: Compute
5. Official MCP servers
We probe official MCP servers every day and record changes in reachability, tool lists and docs. This week:
- FinChina's cross-border reachability changed six times in seven days (blocked and unstable in turn). As of this morning it is unstable.
- Google Maps MCP added a
search_along_routetool on Nov 7, and parameters of several tools changed again on Nov 9. - Salesforce left the list on Nov 4.
A static list that says "has MCP" can't tell an agent whether the server answers today or what its tools look like now.
Source: Open MCP
Get the data from your agent
All of the above is available through the TANDIU MCP server (Streamable HTTP, no authentication for the free tools):
{"mcpServers":{"tandiu":{"url":"https://www.tandiu.cn/mcp"}}}
Useful tools for this week's topics: list_price_changes, get_price_index, list_ai_funding, list_gpu_prices, get_mcp_changes. Setup examples: tandiu-ai/tandiu-mcp.
Public data compiled by TANDIU. Not investment or procurement advice.
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.