Three labs now price their mid-tier flagship at exactly $2 in, $10 out, and the flagship price war of 2026 is the reason. ByteHerald is starting a monthly tracker of frontier model API pricing: GPT-6, Claude, and Gemini, per million tokens, checked against provider pages and current technical coverage. This is the October 2026 baseline. We will re-run it every month, so the history of who cut what, and when, accumulates here instead of scattering across launch blogs.
The October 2026 database
| Model | Input / 1M | Output / 1M | Blended* / 1M |
|---|---|---|---|
| GPT-6 Astra (OpenAI flagship) | $10.00 | $50.00 | $20.00 |
| Claude Fable 5.1 (Anthropic) | $10.00 | $50.00 | $20.00 |
| Claude Opus 5.5 (Anthropic) | $4.00 | $20.00 | $8.00 |
| GPT-6.1 Sol (OpenAI) | $2.00 | $10.00 | $4.00 |
| Claude Sonnet 5.5 (Anthropic) | $2.00 | $10.00 | $4.00 |
| Gemini 4 Argon (Google, introductory) | $2.00 | $10.00 | $4.00 |
| Gemini 3.8 Flash (intro, thru Dec 31) | $0.75 | $3.75 | $1.50 |
| Gemini 3.5 Flash-Lite | $0.30 | $2.50 | $0.85 |
| GPT-6 Luna (OpenAI) | $0.10 | $0.50 | $0.20 |
| Claude Haiku 5.5 (<100K requests) | $0.10 | $0.50 | $0.20 |
*Blended at 750K input + 250K output tokens per million, a common mix for agent workloads.
The price-war timeline: how we got here
| Date | What moved |
|---|---|
| Jul 2026 | Claude Opus 5 ships at $5/$25; four 2026 Opus releases hold one price line |
| Aug 2026 | Anthropic makes Claude Sonnet 5's introductory $2/$10 pricing permanent |
| Sep 22, 2026 | Claude Opus 5.5 cuts to $4/$20, a 20% cut, cache reads down 60% |
| Late Sep 2026 | OpenAI launches GPT-6 Sol at $2/$10 (exactly Sonnet 5's price) and Luna at $0.10/$0.50 |
| ~Oct 7, 2026 | Claude Haiku 5.5 launches at $0.10/$0.50, matching Luna, a 90% reduction |
| Oct 2026 | Gemini 4 Argon launches at introductory $2/$10, joining the $2/$10 cluster |
Three findings
1. The $2/$10 convergence is real-time competitive pricing. GPT-6.1 Sol, Claude Sonnet 5.5, and Gemini 4 Argon all landed on the identical price pair within weeks. Nobody prices like this by accident, the labs are watching each other's pricing pages and matching them the way airlines match fares.
2. The flagship premium is enormous. Astra and Fable 5.1 cost 5x the $4.00 cluster and 100x the budget tiers per blended million tokens. But Anthropic's own estimate is the important nuance: Opus 5.5 finishes tasks in fewer tokens, so it claims typical workloads run about 40% cheaper than the list price suggests. Per-token price is not per-task cost.
3. Introductory prices are windows, not rates. Gemini 4 Argon is explicitly temporary, Google says it rises to $4/$20 with no date given. Gemini 3.8 Flash's $0.75/$3.75 ends January 1, 2027, when it doubles. Haiku 5.5 quintuples past 100K tokens; Luna surcharges past 272K input tokens. The list price is the start of the math, never the end.
How this tracker works
We check list prices against provider pricing pages and current technical coverage at the start of each month, record the date, and keep the history above growing, so a cut that looks like a launch-day stunt reads as what it is: a move in a sequence. We ran a narrower version of this check on October 7, when Mistral Large 4 launched and undercut every Western frontier model, and the same October pricing tables powered our coding-assistant scorecard. Prices change frequently; check the provider's live page before budgeting.
No provider paid for or reviewed this analysis. Rates are list prices excluding caching, batch discounts, tools, regional premiums, and negotiated rates. This compares prices, not equivalent capabilities.
Comments
Post a Comment