Blog · Category
Head-to-head LLM and API provider comparisons from LLM Waves Research. Independent benchmarks, published methodology, no vendor funding.
30 articles
Cursor and Claude Code both start at $20 a month. From each vendor's own pages: team pricing, Anthropic's real $150 to $250 cost, and running both together.
10 min read
Perplexity's $20 plan routes to four outside models, Claude included. Claude's $20 plan draws on none. Compare pricing, context windows and agents.
9 min read
Claude Haiku 4.5 vs Sonnet 5 compared on pricing, a five times narrower context window, and effort levels, from Anthropic's own docs, September 22, 2026.
10 min read
Claude Sonnet 5 vs Opus 5, from Anthropic's own docs: pricing, all five effort levels, cache limits, and the benchmark gap neither model's page ever closes.
11 min read
DeepSeek-V4-Pro prices output tokens up to 25 times below ChatGPT's priciest tier, yet above OpenAI's own cheapest model. Compare every tier plus uptime.
11 min read
Claude vs ChatGPT vs Gemini: all three cost $20 a month, but each also sells a $100 and $200 tier. See all six pricing tiers before you choose.
11 min read
Claude Opus 5 and Gemini 3.1 Pro compared on API pricing, benchmark sourcing, and output limits, using each vendor's own pricing page and model card, 2026.
10 min read
Gemini API pricing, collected from Google's own docs: current Flash and Pro rates, the January 2027 rate increase, and how grounding is billed by search.
12 min read
Grok 4.6 and Claude Opus 5 API pricing, coding benchmarks, and metered tool fees compared from xAI's and Anthropic's own pages, collected September 10, 2026.
11 min read
DeepSeek API pricing after the August 16 peak/off-peak split: $0.66 to $3.96 per million tokens, the 7 UTC hours that double it, and who actually pays peak.
10 min read
DeepSeek and Claude were compared on September 2026 pricing, SWE-bench Verified and Pro scores, data residency, and self-hosting cost, sourced from each vendor.
10 min read
Claude API pricing for Fable 5.1, Opus 5, Sonnet 5, and Haiku 4.5: the cancelled hike, tool-use overhead, and thinking-token cost, all collected on 2026-09-08.
10 min read
Aggregated pricing and SWE-bench Verified data across 12 coding LLMs. Claude Opus 5 leads accuracy at 96 percent, DeepSeek prices tasks 90 times cheaper.
17 min read
LLM API pricing compared for 15 models from six vendors: output costs $0.50 to $50 per 1M tokens, a 100x spread, from vendor pricing pages on October 1, 2026.
13 min read
DeepInfra prices DeepSeek-V4-Flash-0731 72% below DeepSeek's own direct rate. Six routes compared on cost, cache discounts, and compliance, dated 2026-09-02.
13 min read
13 Llama hosting providers compared on pricing, context windows, compliance and access. Prices split 73% on Llama 3.3 70B output, collected September 2026.
16 min read
Kimi K2.6 leads open source LLMs for coding at 80.2% SWE-bench Verified, ahead of GLM-4.7 and Qwen. Licenses, self-host costs, and agent support compared.
17 min read
Aggregated from vendor cards: six 2026 open weight models compared on license, VRAM tier, and GPQA Diamond score, so you can download the right quantization.
15 min read
OpenRouter lists the same per token price providers list directly, then adds a 5.5% credit fee or 5% BYOK fee above $25,000 monthly. Collected 2026-08-27.
10 min read
OpenRouter free tier limits, aggregated from OpenRouter's own documentation: 20 requests a minute, 50 or 1,000 a day depending on a one-time $10 purchase.
10 min read
Aggregated pricing and speed data for 13 LLM API providers, compared on 2026-08-26: Groq measures fastest, DeepInfra prices output 34% below Together AI.åå
17 min read
OpenRouter free models rotate, but the 20 requests a minute cap and the 50 to 1,000 daily ceiling do not. Collected from OpenRouter's own docs, August 2026.
11 min read
Novita AI and Hyperbolic were compared on published pricing structure, model catalog size, and GPU hardware, using vendor documentation retrieved August 26, 2026.
10 min read
Aggregated comparison of Fireworks AI and DeepInfra: per-token pricing, GPU rates, caching, batch discounts, and rate limits, all sourced August 25, 2026
13 min read
Cerebras and Groq compared on gpt-oss-120b speed, price, and context window, each figure sourced, plus the Nvidia deal behind Groq's 2026 valuation reset.
9 min read
Aggregated pricing and rate-limit data compare Groq API and OpenAI API: Groq's cheapest model halves OpenAI's output token price.
8 min read
Basis: aggregated. Same model, different providers: three models, 46 endpoints show up to a 29-point accuracy gap and a 34x speed gap, retrieved 2026-08-21.
9 min read
Time to first token, or TTFT, explained: the formula, what changes it, and how ten endpoints compare, 0.16 to 1.12 seconds, aggregated on August 19, 2026.
9 min read
Groq vs Grok: Groq sells AI inference cloud capacity, and Grok is the SpaceXAI chatbot. Ownership, Grok 4.7 and GPT-OSS prices, and trademark history.
10 min read
Best speech to text models: MAI-Transcribe-2 and Grok Voice Transcribe 2.0 get about 2 words wrong in 100 for $0.10 an hour. Best free pick: Qwen3-ASR.
9 min read