Three frontier models launched within 48 hours at prices within a factor of three. A side-by-side on cost, context, output limits and published benchmarks, and how to actually choose.
Travis Johnson
September 23, 2026 · 4 min
Anthropic's Opus 5.5 landed at $4/$20 per million tokens, a 20% cut against Opus 5, with a 60% cheaper cache read. What changed, and when Fable 5.1 is still worth the money.
Travis Johnson
September 23, 2026 · 3 min
xAI's Grok 4.7 shipped at the same list price as 4.6 with better scores across the board. The spec that actually stands out is an output limit three and a half times any rival's.
Travis Johnson
September 22, 2026 · 3 min
Qwen 3.8 Omni Flash takes text, image, audio and video for $0.15 per million input tokens. A tour of the budget tier as it stands in September 2026, with verified prices.
Travis Johnson
September 12, 2026 · 3 min
Two labs priced their flagship at exactly $10/$50 three days apart. What that convergence means, and when paying double the tier below is actually justified.
Travis Johnson
September 5, 2026 · 4 min
GLM 5.3 Flash offers a 1.31M context window, larger than any frontier model, at a fraction of the price. Why specifications are not capability, and what long context is actually for.
Travis Johnson
August 28, 2026 · 4 min
Anthropic's Opus 5 launched on 24 July at $5/$25 per million tokens, outscoring the more expensive Fable 5 on both published indices. When the top of a lineup stops being the best buy.
Travis Johnson
July 29, 2026 · 4 min
Context windows converged on a million tokens. Output ceilings did not: they run from 16,000 to over 943,000, and the largest come from labs outside the frontier three.
Travis Johnson
July 21, 2026 · 4 min
We tested factual accuracy on a standardized question set across 8 major AI models. The hallucination rates (and the types of errors each model makes) differ significantly and have real implications for how you should use each.
Travis Johnson
April 2, 2026 · 13 min
No single image model dominates every visual style. We map the top AI image generators to specific aesthetic categories: photorealism, concept art, illustration, product photography, and more.
Travis Johnson
March 20, 2026 · 10 min
Flux has largely displaced Stable Diffusion as the open-weight image generation standard. We compare both on quality, customization, and deployment to help you choose the right open model.
Travis Johnson
March 4, 2026 · 10 min
We ran identical prompts through the three leading AI image generators across photorealistic, artistic, and illustration styles. The results reveal distinct strengths that make each model best for different creative work.
Travis Johnson
February 24, 2026 · 11 min
Gemini 2.5 Ultra leads on long-context tasks, multimodal reasoning, and Google Workspace integration. We tested it thoroughly and compare it to GPT-5 and Claude 4 across 10 task categories.
Travis Johnson
January 31, 2026 · 12 min
A full pricing matrix for 30+ AI models: input cost, output cost, and a value score combining price with benchmark performance. Essential reading for developers choosing models for production applications.
Travis Johnson
August 28, 2025 · 10 min
Reasoning models think before they answer, and the quality difference on complex tasks is substantial. We compared o3, Gemini 2.0 Thinking, and Claude Extended Thinking on math, logic, and multi-step problems.
Travis Johnson
August 20, 2025 · 13 min
Speed matters for interactive AI applications. We benchmarked tokens per second and first-token latency across 15+ models to rank the fastest LLMs and explain when to choose speed over quality.
Travis Johnson
August 12, 2025 · 9 min
Context windows range from 8K to 2 million tokens. We tested real performance at different lengths (not just advertised limits) to find which models actually deliver on their long-context promises.
Travis Johnson
August 4, 2025 · 10 min
A comprehensive, regularly updated benchmark table for 20+ major AI models across MMLU, HumanEval, MATH, MT-Bench, and GPQA, with plain-English explanations of what each score actually means.
Travis Johnson
July 19, 2025 · 14 min
We tested 6 models on academic papers, legal documents, news articles, and business reports. The results reveal significant differences in compression quality, hallucination rate, and key-point retention.
Travis Johnson
July 11, 2025 · 10 min
The open-weight AI landscape has never been more competitive. We compared Qwen 2.5, DeepSeek V3, and Llama 4 across performance, licensing, and deployment to find the best open model for each use case.
Travis Johnson
July 3, 2025 · 12 min
Long-context handling, citation accuracy, and multi-source synthesis are where AI models diverge most. We tested 6 models on real research tasks to find the best AI research assistant.
Travis Johnson
June 25, 2025 · 12 min
We compared GPT-4o, Claude 3.5 Sonnet, Gemini, and 4 others on blog posts, emails, marketing copy, creative fiction, and technical documentation to find the best AI writing assistant.
Travis Johnson
June 17, 2025 · 11 min
Meta's Llama 4 is the most capable open-weight model yet. We benchmarked it against GPT-4o and Claude to quantify the capability gap, and found it smaller than most people expect.
Travis Johnson
June 9, 2025 · 12 min
Grok 3 brings real-time X/Twitter data and a distinct personality. We tested it against GPT-4o on reasoning, humor, coding, and factual accuracy to find out if it's a genuine ChatGPT rival.
Travis Johnson
May 16, 2025 · 10 min
A rigorous side-by-side comparison across 8 task categories (coding, writing, summarization, math, creative tasks, reasoning, instruction following, and factual accuracy) with a use-case recommendation matrix.
Travis Johnson
April 30, 2025 · 13 min
We tested GPT-4o, Claude Sonnet, Gemini 2.0, DeepSeek Coder, and 6 others on real coding tasks: debugging, architecture, code review, and documentation. The rankings might surprise you.
Travis Johnson
April 8, 2025 · 14 min
We ran 50 real-world prompts through GPT-4o, Claude Opus, and Gemini Pro simultaneously. Here's what we found, and why the "best" model depends entirely on your use case.
Travis Johnson
March 15, 2025 · 12 min
We publish model comparisons, prompt guides, and AI news. No spam, unsubscribe anytime.
Try Deepest free