What Entered the Catalog
| Model | Rate ($/1M in, out) | Type | Why it matters |
|---|---|---|---|
| DeepSeek V4 Flash Vision Exp | $0.44 / $1.32 | Multimodal budget | Image understanding at the cheapest serious rate in the market |
| Qwen3.8 27B | $0.42 / $3.00 | Open-weight balanced | Dense 27B with 1M context — strong CJK + coding value |
| Muse Spark 1.2 Contributor | $0.10 / $0.20 | Near-free reasoning | Meta's contributor tier — but prompts may train Meta products |
| Tencent Hy-MT2-30B-A3B | $0.074 / $0.295 | Translation specialist | 33 language pairs at the lowest per-token price in the catalog |
The pattern: the new supply is all at the bottom of the price curve. Frontier pricing barely moved this month while the budget tier added four more weapons — the quality-per-dollar staircase keeps getting steeper.
What Got Cheaper (and What Didn't)
Promotional listings: GPT-5.6 Sol and Gemini 3.7 Flash
OpenRouter listed Sol at $2/$10 (list $5/$30) and 3.7 Flash at $0.75/$3.75 (list $1.50/$6) in late August. If you buy through OpenRouter, promos like these are real savings while they last — but they are not official list prices, and this catalog only records official rates.
Off-peak DeepSeek: still the floor
V4 Flash off-peak at $0.22/$0.66 with $0.014 cached input — the cheapest serious production rate. Peak/off-peak routing now moves real money: shifting 60% of a pipeline off-peak halves that model's line.
Unchanged: frontier list prices
Sol, Opus 5, Fable 5 and Gemini Pro list prices did not move officially in August. The value story is the budget tier climbing, not the frontier falling.
What Disappeared
Two free routes left OpenRouter within a week of each other:
- Ox Alpha (stealth/ox-alpha). Removed 2026-08-28 — the free 1M-context preview ended without a provider reveal. Full timeline and fallbacks in the removal analysis.
- Dots3-Note Preview. Gone ahead of its announced September 30 sunset — free open-weight previews are ending early, not late.
Read the full story in Ox Alpha is gone from OpenRouter, and the economics of depending on free routes in why free AI tiers aren't free.
What This Means for Your Stack
Re-run your routing mix monthly — the August changes moved relative rankings, not just prices.
If you bought Sol or 3.7 Flash through OpenRouter promos, price the post-promo list rates before committing capacity.
The new budget entrants (Qwen3.8 27B, Hy-MT2) are worth an eval if your workload is CJK-heavy or translation-heavy.
Anything still depending on a free preview route needs a paid fallback this week.
Frequently Asked Questions
What changed in AI pricing in August 2026?
Four storylines: the catalog grew (Qwen 3.8 27B, Muse Spark 1.2 Contributor, Tencent Hy-MT2, DeepSeek V4 Flash Vision Exp), free previews ended (Ox Alpha, Dots3-Note), budget-tier prices kept falling on OpenRouter (GPT-5.6 Sol and Gemini 3.7 Flash listed well below list during promos), and off-peak DeepSeek stayed the cheapest serious production option.
Did flagship prices actually drop in August?
On OpenRouter, yes — GPT-5.6 Sol listed at $2/$10 and Gemini 3.7 Flash at $0.75/$3.75 during late-August promos, roughly half of list. We treat those as promotional listings until the official pages confirm; the catalog re-verification pass (2026-08-28) is the source of truth.
What is the cheapest serious model right now?
DeepSeek V4 Flash off-peak at $0.22/$0.66 with a $0.014 cache-hit rate. GPT-5.6 Luna ($0.20/$1.20) and Gemini 3.5 Flash-Lite ($0.30/$2.50) are the close rivals for high-volume traffic.
Should I re-negotiate my model stack after these changes?
Yes — the August moves changed relative rankings, not just absolute prices. The routing playbook's quarterly re-audit step is exactly for this: re-run your mix, re-check the benchmark hub, and update routing rules.
Where can I see all current prices?
The model catalog lists every verified rate with source links and verification dates, and the pricing trends page tracks the history. Every price display on this site carries its verification timestamp.