Loading Frontier AI Ledger...
10 AI models compared: Claude Fable 5, GPT-5.6, Grok 4.5, Muse Spark 1.1, Gemini 3.1 Pro, DeepSeek V4-Pro, Qwen 3.7 Max, Kimi K2.6, Mistral Large 3, and Llama 4 — benchmarks, pricing, funding, and 47 tracked model releases across 8 years.
Across 5 frontier labs
OpenAI + Anthropic + xAI
$1.25 to $50 per M tokens
Biggest transparency gap tracked
Built by Trace Cohen · Free to use · Data updated 2026-07-09 · Reach out at t@nyvp.com
Between June 9 and July 9, 2026, Meta, xAI, OpenAI, and Anthropic each shipped a new flagship model within a single 30-day window — and two of them got pulled back by governments before the dust settled. Claude Fable 5 was suspended for 22 days under US export controls; GPT-5.6 was gated for 12 days pending a government safety review. Widen the lens past those four labs and the picture gets more interesting: Google's Gemini 3.1 Pro and DeepSeek's V4-Pro actually top the published benchmark scores outright, while OpenAI discloses zero of eight tracked benchmarks for GPT-5.6 at launch — the biggest transparency gap of any lab tracked. Pricing spans a 40x range, from Meta's $1.25 per million input tokens to Anthropic's $50 per million output tokens. Behind the models sits over $2 trillion in combined valuation across OpenAI ($852B), Anthropic ($965B), and xAI ($230B pre-SpaceX-merger) — this dashboard tracks the full funding history for all three, plus a 47-release model family tree spanning GPT-1 through GPT-5.6, Claude 1 through Fable 5, and Gemini 1.0 through 3.1.
Claude Fable 5 tops the Artificial Analysis Intelligence Index at 60 and leads on independently-verified SWE-bench Verified (95%), but Google's Gemini 3.1 Pro scores highest on raw GPQA Diamond (94.3) and ARC-AGI-2 (77.1). Grok 4.5 wins on price-to-performance, and DeepSeek V4-Pro is the cheapest frontier-class option at $0.435 per million input tokens.
Per million tokens (input/output): Meta's Muse Spark 1.1 is cheapest at $1.25/$4.25, xAI's Grok 4.5 runs $2.00/$6.00, OpenAI's GPT-5.6 Sol is $5.00/$30.00, and Anthropic's Claude Fable 5 tops out at $10.00/$50.00 — a 40x spread between the cheapest and priciest frontier model tracked.
OpenAI has raised roughly $180B across 14 rounds to reach an $852B valuation. Anthropic has raised about $132B across 18 rounds to hit $965B. xAI raised roughly $45B across 9 rounds to a $230B Series E valuation before merging into SpaceX. Every round for all three is tracked with dates, amounts, and investors.
OpenAI's GPT-5.6 discloses 0 of 8 tracked benchmarks at launch — no GPQA, SWE-bench, AIME, ARC-AGI-2, or Humanity's Last Exam scores published. Meta discloses just 1 of 8. Anthropic (8/8) and DeepSeek (6/8) are the most transparent labs tracked, though several of Anthropic's figures are third-party sourced rather than self-reported.
Claude Fable 5 leads on independently-verified benchmarks (95% SWE-bench Verified) and tops the Artificial Analysis Intelligence Index at 60, but Gemini 3.1 Pro scores highest on raw GPQA Diamond (94.3) and ARC-AGI-2 (77.1). There's no single “best” model — Grok 4.5 wins on price-to-performance, and DeepSeek V4-Pro is the cheapest frontier-class option at $0.435 per million input tokens.
Pricing per million tokens (input/output): Meta's Muse Spark 1.1 is cheapest at $1.25/$4.25, xAI's Grok 4.5 is $2.00/$6.00, OpenAI's GPT-5.6 Sol is $5.00/$30.00, and Anthropic's Claude Fable 5 is the most expensive at $10.00/$50.00 — a 40x spread between the cheapest and priciest frontier model.
OpenAI has raised roughly $180B across 14 rounds, reaching an $852B valuation after its $122B round in March 2026. Anthropic has raised about $132B across 18 rounds, hitting $965B after a $65B Series H in May 2026. xAI has raised roughly $45B across 9 rounds before its $230B Series E and subsequent SpaceX merger.
The US Commerce Department forced Anthropic to suspend Claude Fable 5 and Mythos 5 globally for 22 days, from June 12 to July 1, 2026, after a jailbreak reportedly let Mythos-tier cyberdefense capability leak through Fable 5's guardrails. Anthropic argued the same exploit could affect competing models with no equivalent restriction, and restored the model with an added safety filter.
OpenAI's GPT-5.6 discloses 0 of 8 tracked benchmarks at launch — no GPQA, SWE-bench, AIME, ARC-AGI-2, or Humanity's Last Exam scores were published. Meta's Muse Spark 1.1 discloses just 1 of 8. By contrast, Anthropic discloses 8 of 8 (though some figures are third-party sourced) and DeepSeek discloses 6 of 8, making them the most transparent frontier labs tracked.