Naderu Weekly · Episode 16
Gated is the price
Gemini 4 Carbon lands on the API with a 1 December doubling, and GPT-6 Cyber launches at $20/$100 behind Daybreak Red — two launches, two gates, one arithmetic. Plus the Image 2.6 pair, a trusted-access-only Rosalind, and a Hy4 release whose pricing we could only find second-hand.
Watch on YouTube — 6:08. Every figure below is linked to where we got it. If you would rather have it arrive than remember to check, subscribe to the weekly report — one email a week, one click to leave.
Last week's open question first: who is actually hosting K3.1 Pro? As far as we could confirm on vendor pages this week, Moonshot alone — no second host cleared the sourcing bar, so the board data file still serves it from one place, and we are not airing a host we cannot cite.
Every gate this week has a date on it
| Date | What changes |
|---|---|
| 30 Oct | GPT-6 Rosalind released, trusted-access Research only |
| 4 Nov | Tencent Hy4 full release plus the GPT-Image-2.6 pair |
| 5 Nov | GPT-6 Cyber, Daybreak Red gated from birth |
| 21 Nov | GPT-5.6 Sol promo rate ends |
| 1 Dec | Carbon and Argon double to $4 / $20 |
| 31 Dec | Hy4 TokenHub half-price ends — second-hand, see below |
Three of those are doors, three are prices going up, and one is a release with no citable price at all.
Carbon clears the three-fact gate
Carbon sat in the board's deferred ledger for one reason: no model id, no context window,
no API pricing page. All three now resolve on Google's own model page:
gemini-4-carbon, a 1,048,576-token unified input-plus-output window, 65,536
default output with long decode to a full million.
Beside them, the price on Google's pricing page: $2 in / $10 out through 30 November, $4 / $20 from 1 December. Paid tier only — no free tier — and implicit caching only, so there are no explicit caches to manage and no cache-storage line to forget.
One run, six prices
Ten million input tokens and two million out, at list rates, uncached, with every request under any long-context step — priced off the board data file:
| Same workload | Job cost |
|---|---|
| Kimi K3.1 Pro Highspeed ($4 / $20) | $80.00 |
| Argon and Carbon from 1 December ($4 / $20) | $80.00 |
| Kimi K3.1 Pro ($2 / $10) | $40.00 |
| Argon and Carbon until 30 November ($2 / $10) | $40.00 |
| GPT-6.1 Terra ($1 / $5) | $20.00 |
| Mistral Large 4 ($0.70 / $2.10) | $11.20 |
The Terra and Kimi rates come off OpenAI's pricing page and Moonshot's. Large 4 at $11.20 is still the cheapest full run on this chart, with Terra at $20 the cheapest route that asks for no licence discussion at all.
Cyber: priced for the vetted few
Every capability number here is OpenAI's own reporting: ExploitGym 48.5%, up from 32.1% for GPT-5.6 Cyber. 948 privilege-escalation exploits developed in Linux, 81 previously unknown high-severity Firefox vulnerabilities identified. In-scope avoidance 99.93% without the deployment prompt, 100% with it.
The gate first, from the Cyber model page: Daybreak Red approval, API and Codex only — no batch, no fine-tuning. Then the price: $20 in, $2 cached, $100 out — $40, $4 and $150 once a request passes 272,000 input tokens. Cache writes are $25 for thirty minutes, $50 for twelve hours.
| Same 10M-in, 2M-out run | Job cost |
|---|---|
| GPT-6 Cyber, long context ($40 / $150) | $700.00 |
| GPT-6 Cyber, short context ($20 / $100) | $400.00 |
| Kimi K3.1 Pro ($2 / $10) | $40.00 |
| GPT-6.1 Terra ($1 / $5) | $20.00 |
The model is twenty times Terra per token — and the approval is the part no budget line can fix.
Also shipped, briefly
GPT-6 Rosalind arrived 30 October — biology on Astra weights, trusted-access Research only, in Work, Codex and the API — per OpenAI's changelog. The GPT-Image-2.6 pair shipped 4 November: Sunburst lists $5 text in, $8 image in and $30 image out on its model page — generation and editing, outside a text-token comparison, so noted and not rowed.
And one item said second-hand, out loud. Tencent's full Hy4 released 4 November — 770B mixture-of-experts, one-million window, vision and computer use — at 6 RMB in and 18 out, half-price on TokenHub through 31 December, per Model Price Watch's Hy4 page. We could not find those numbers on a Tencent page we can cite, so they are not a board row until we can.
Three moves for Monday
1. Budget the 1 December doubling
Carbon and Argon both go $2/$10 to $4/$20. Pin the rate your launch actually runs on before the thirtieth.
2. Treat Cyber as gated, not as priced
$20 in is not the barrier; Daybreak Red approval is. Without it there is no row in your plan — keep Terra or Sonnet as the route.
3. Keep Hy4 second-hand until a table exists
Record the RMB figures as a dated option, not an input — re-check the first-party page before you spend.
Also on the board this week
The board's last full refresh is 5 November (74 rows). This week's deltas for the next pass: Carbon's model page, window and pricing row now clear the three-fact gate; Cyber arrives with a $20/$100 short-context row behind Daybreak Red; the Image 2.6 pair is noted and not rowed; Hy4 stays deferred with no citable first-party table. Every figure above is on the board or in its deferred ledger.
Sources
- Naderu model board and its data file.
- Gemini 4 Carbon model page; Google pricing.
- Introducing GPT-6 Cyber; Cyber model page; OpenAI changelog; Image 2.6 Sunburst page; OpenAI pricing.
- Moonshot pricing; Mistral pricing.
- Model Price Watch Hy4 page — second-hand, said out loud in the episode.