Price Changelog
Every model addition and price change we track, with official sources. Subscribe via RSS. Upcoming changes are flagged.
Gemini 3.x promo expires: prices double (+100%) +100%
Per Google's official pricing page, promotional rates for Gemini 3.6/3.7/3.8 Flash double on Jan 1, 2027: $0.75/$3.75 → $1.50/$7.50. (0.75/3.75 → 1.50/7.50 per 1M)
7 models removed from OpenAI pricing page — major cleanup
gpt-5.6-sol (promo ended early), 5.6-cyber, 6-sol, 5.3-codex, rosalind-research, rosalind-discovery, image-2.5 all removed. GPT-5.x generation fully retired.
Fable 5.2 & Mythos 5.2 added ($8/$40) — 20% price cut
Both upgraded from 5.1 with 20% reduction ($10/$50 → $8/$40). Anthropic's second price cut of 2026, while Google prepares to double prices.
GPT-6 Cyber added ($20/$100) — 60% pricier than 5.6-Cyber
New GPT-6 generation Cyber model. Input up from $12.50 to $20, output from $75 to $100.
GPT-6 Astra Law added ($10/$50) — OpenAI GPT-6 family complete
Legal specialization variant, same pricing as base Astra. Promotional through Dec 31. All 7 GPT-6 variants now tracked.
GPT-6 Rosalind added ($15/$75) — new ultra-premium tier
Highest-priced GPT model, 50% above Astra. Cached $1.50. Detected by monitoring.
GPT-6.1 Terra added ($1/$5)
New mid-tier model between Sol and Luna. Cached $0.05. User discovered before monitoring alert.
Haiku 5.5 long-context prices corrected (input/output were reversed)
Mathematical proof: cache_write_5min = 1.25×input → $0.625/1.25 = $0.50 input (not $2.50). Output is $2.50 (not $1.00). Also added gpt-4o-mini-transcribe ($1.25/$5.00) to changelog (was missing).
Medium-priority models + comparison pages expansion
+6 models (Opus 4.1, Flash Image, Pro Image, Transcribe×3) and +3 comparison pages (1 removed: veo-vs-sora 500, 1 replaced) (Grok 4.6vs4.7, Realtime vs Mini, Haiku5.5 vs Luna, Veo3.1 vs Sora2). Now 61 models.
Major expansion: +8 models (Realtime/Image/Live/Chat-Latest/Fable5/Mythos5/Veo3.1)
Added GPT Realtime 2.1($32/$64) + Mini($10/$20), GPT Image 2.5($8/$30), Chat Latest($5/$30), GPT Live 1($0.05/min), Fable 5, Mythos 5, Veo 3.1($0.40/sec). Now 55 models across 6 providers.
Nano Banana 2.1 added ($0.50 in / $30 image out)
Google's image generation model now on the pricing page. Text input $0.50/1M, image output $30/1M. Previously only referenced in CSS — now has full pricing data.
Rosalind Discovery added + GPT-5.6 Sol promo expiry revealed (Nov 21)
New model gpt-rosalind-discovery ($5/$25, same as Research). GPT-5.6 Sol promotional pricing 'available at least through November 21, 2026' — potentially a third price increase event in the Nov-Jan window.
Gemini 4 Argon pricing published: $2/$10 promo, doubles Dec 1 (+100%) first published
Google published Argon pricing after initially showing 'Not available'. Introductory rates: $2 input / $10 output / $0.10 cache through Nov 30. All rates double Dec 1 — one month BEFORE the Gemini 3.x Jan 1 doubling. Two consecutive price increases in 31 days. (TBA → 2.00/10.00 per 1M)
Gemini 4 Argon added (no public pricing yet)
Flagship reasoning model, enterprise-only at launch. Official pricing page shows 'Not available'. Will update when Google publishes rates.
Claude Haiku 5.5 added ($0.10/$0.50 — cheapest Claude ever)
Ultra-budget tier with long-context split at 100K tokens (>100K: $2.50/$1.00). 10x cheaper than Haiku 4.5.
Comparison pages launched (+3)
Head-to-head cost comparisons with workload calculator: GPT-6 Sol vs 5.6 Sol, GPT-6 Astra vs Luna (100x gap), Gemini 3.8 Flash vs 3.1 Pro — triggered by comparison queries forming in search autocomplete.
Gemini 3.7 Flash removed from official pricing page
Google delisted 3.7 Flash from ai.google.dev pricing (3.8 and 3.6 remain at $0.75/$3.75 promo). Likely retired — page kept with historical rates and deprecation banner.
Claude Sonnet 5.5 added ($2/$10)
Newest Sonnet generation at unchanged Sonnet 5 rates — capability upgrade at no extra cost.
GPT-6.1 Sol added ($2/$10)
6.1 generation of the Sol workhorse, same headline rates as 6.0: $2/$10, cached $0.10, batch $1/$5, long-context 2x.
Qwen hosted pricing added (+4 models)
Qwen3.8-27B (¥3/¥12), 3.6-27B, 3.5-122B and 3.5-35B — hosted rates on SiliconFlow, CNY per 1M tokens. Llama deferred (not listed on the captured source).
Qwen prices switched to international USD rates
Replaced CN-platform CNY rates with SiliconFlow international USD pricing: Qwen3.8-2.4T-A95B $2/$6 (cached $0.25), 3.6-27B $0.3/$3.2, 3.5-122B $0.26/$2.08, 3.6-35B $0.24/$1.8. CN platform still bills CNY — noted on each page.
Grok lineup added (+5 models)
Grok 4.7/4.6 ($2/$6), 4.5 ($2/$6, cheaper cache), 4.3 ($1.25/$2.50) and Build 0.1 ($1/$2), with cached-read and 2x long-context rates from docs.x.ai.
DeepSeek added (Flash + V4 Pro)
Flash $0.30/$1.20 and V4 Pro $1.32/$3.96 at peak, with off-peak 50% discounts and cache-hit input at $0.006/$0.044.
GPT-6 family added (+7 models)
GPT-6 Astra ($10/$50), Sol ($2/$10), Luna ($0.10/$0.50), plus GPT-5.6 Sol/Cyber, 5.3 Codex and Rosalind Research. Batch = 50%, Priority = 2x.
Public API + llms.txt
Launched free JSON API (/api/v1/prices), llms.txt index and FAQ/WebSite structured data.
LLMPrices.org launched
Site launched with Gemini 3.8/3.7/3.6 Flash and Claude Opus 5, Sonnet 5, Opus 4, Sonnet 4 pricing, tracked from official docs.
Anthropic lineup completed (+9 models)
Added Opus 5.5 ($4/$20), Fable 5.1 and Mythos 5.1 ($10/$50), Haiku 4.5 ($1/$5), and the Opus 4.5–4.8 / Sonnet 4.5–4.6 line.
Opus 4.5–4.8 repriced: $15/$75 → $5/$25 (-67%) -67%
Per Anthropic docs, the late Opus 4 generation now lists at $5/$25 per MTok — a 67% cut vs Opus 4/4.1 launch pricing. (15/75 → 5/25 per 1M)
Gemini side expanded (+5 models)
Added 3.1 Pro (tiered ≤/>200k), 3.5/3.1 Flash-Lite, 3.8 Flash TTS and Omni Flash (video ≈$0.10/sec).