TokenBlog
Model releases, pricing breakdowns, and practical guides for developers.
Every major LLM API now knocks about 90% off the tokens you cache. The bill you don't see…
Cached input reads have quietly standardized near a tenth of the base price across OpenAI, Anthropic, and Google, and DeepSeek goes further at under 1%. The real spread is on the write side: Anthropic…
American teams now route a third of their tokens to Chinese models to save money. The plot…
US companies have made up more than 30% of Chinese-model token traffic on OpenRouter every week since February, and the reason is the…
8 min read
For two years the story out of China was that prices only fall. Moonshot just tripled its…
Kimi K3 lists at $3/$15 per million tokens against K2.6's $0.95/$4.00, a 3.16x input and 3.75x output increase under three months apart. An…
9 min read
A Kuaishou team just priced 94 percent of Opus 4.8's coding score at a seventh of the…
Kwaipilot's KAT-Coder-Pro v2.5 lists at $0.74/$2.96 per million tokens, with a cheaper Air tier at $0.15/$0.60. On SWE-Bench Pro it scores…
8 min read
Claude runs five times apart from its cheapest tier to its priciest. The capability gap is…
Anthropic's lineup runs $1/$5 for Haiku 4.5, $3/$15 for Sonnet 5, and $5/$25 for Opus 4.8 - a clean 5x from floor to ceiling. Priced by the…
8 min read
Grok 4.5 sells the cheapest token, Fable 5 sells the smartest one, and neither is the…
GPT-5.6 Sol ($5/$30), Claude Fable 5 ($10/$50), and Grok 4.5 ($2/$6) span five times over on price but bunch near the top on capability…
9 min read
Claude Fable 5 is back on subscriptions, but not the way it left. Included access is gone…
Anthropic pulled Fable 5 from Pro, Max, and Team plans in June on capacity, then brought it back behind metered usage credits at its full…
8 min read
The first model Meta ever charged for is priced to win agent work, not coding. At $1.25…
Meta opened its first paid API on July 9 with Muse Spark 1.1 at $1.25 input and $4.25 output per million tokens, roughly a quarter of what…
8 min read
Grok 4.5 undercuts Opus 4.8 by two-thirds and burns fewer tokens getting there. The…
xAI shipped Grok 4.5 on July 8 at $2 input and $6 output per million tokens. Against Opus 4.8, Fable 5, and GPT-5.6 Sol it runs 68 to 84…
8 min read
Meituan's LongCat-2.0 costs 75 cents a million and tops the open coder benchmarks. You…
LongCat-2.0 went public June 30 at $0.75 input and $2.95 output per million tokens, with a launch promo cutting that to $0.30/$1.20. It…
8 min read
OpenAI built a chip to make inference cheaper. That is a different thing from making your…
Jalapeño, OpenAI's first custom inference processor with Broadcom, landed June 24 alongside Google's Ironwood TPUs and AWS Trainium 3. The…
8 min read
OpenAI split GPT-5.6 into three tiers and priced the flagship exactly where GPT-5.5…
GPT-5.6 arrived June 26 as Sol, Terra, and Luna at $5/$30, $2.50/$15, and $1/$6 per million tokens. Sol, the flagship, costs to the cent…
7 min read
Claude Sonnet 5 lands level with Opus 4.8 on an independent benchmark and asks 60% less…
Anthropic shipped Sonnet 5 on June 30 at $2/$10 per million tokens, an introductory rate that runs through August 31 before rising to…
8 min read
DeepSeek V4 will charge double during its busiest hours. Whether you ever pay the…
From the mid-July V4 launch, DeepSeek doubles its API rates during two daily windows - 9am to 12pm and 2pm to 6pm Beijing time. It is the…
7 min read
The US gated its two newest flagships in a day. The models you can still call are open…
On June 26 the US held OpenAI's GPT-5.6 to government-approved partners and let Anthropic re-open Claude Mythos 5 to a short list of trusted…
8 min read
Doubao 2.1 Pro undercuts Claude Opus by 80%. The price is real, but it is quoted in yuan…
ByteDance launched Seed 2.1 (Doubao 2.1) on June 24 with Pro and Turbo tiers priced only in yuan: roughly $0.88 input and $4.41 output per…
7 min read
Nex-N2-Pro scores like a frontier coder and bills like a budget one. The free weights are…
A little-known lab, Nex AGI, shipped a 397B open-weight model that posts 80.8 on SWE-Bench Verified and costs about a tenth of GPT-5.5…
7 min read
Google promised Gemini 3.5 Pro in June. It is late, has no price, and no benchmarks.
Google teased Gemini 3.5 Pro at I/O on May 19 with a 2M context window and Deep Think, then went quiet. No rate card, no benchmark sheet…
7 min read
Claude won't bill you for a refusal. A fallback credit stops you paying twice.
Claude Fable 5 has been dark for twelve days, but the billing rules it shipped with outlast it. Anthropic does not charge you for a refused…
7 min read
Three frontier coders now cost under a dollar a million. The cheapest sticker is not the…
DeepSeek V4-Pro, Qwen3.7 Plus, and MiniMax M3 all undercut $0.60/M input. Sort by input price and Qwen wins; run an actual coding month and…
8 min read
GLM-5.2 costs a fifth of GPT-5.5 and says it codes better. One of those claims you can…
Z.ai's open-weight GLM-5.2 undercuts GPT-5.5 by roughly 5x per token and self-reports a coding-benchmark win on top. The price is on the…
7 min read
GPT-5.6 is probably days away. Here is what the leaks confirm and what your bill might do.
A gpt-5.6 ID leaked from OpenAI's own Codex backend, the chief scientist reportedly called it a meaningful improvement, and the release-date…
7 min read
Anthropic charges double and keeps winning. Now OpenAI might blink on price.
OpenAI and Anthropic filed to go public a week apart, then OpenAI was reported to be weighing token price cuts to slow Anthropic down. The…
7 min read
Claude Fable 5 lasted three days. Every alternative is cheaper.
Anthropic disabled Claude Fable 5 and Mythos 5 on June 12, 2026 under a US export-control order, three days after launch. Fable 5 was the…
7 min read
GLM 5.2 and Kimi K2.7 Code shipped five days apart. You still cannot benchmark one against…
Two Chinese labs put out open-weight coding models in the same week. Kimi K2.7 Code lists at $0.95/$4.00 per million tokens, GLM 5.2 at…
6 min read