llm-token-cost-analyses.md

LLM Token Cost Over Time — Key Analyses

Source	Date	Headline Finding	Methodology	Key Caveats
Epoch AI ¹	Mar 2025	Price decline of 9×–900× per year depending on benchmark; median ~50× (or ~200× post-Jan 2024)	Tracked cheapest model to hit specific benchmark milestones across 6 benchmarks; 3:1 input/output weighted average; excluded reasoning models	Benchmarks don't capture all useful capabilities; possible benchmark overfitting; fastest trends are recent and may not persist
a16z ²	Nov 2024	~10× cost reduction per year at equivalent performance	MMLU scores + historical pricing from Internet Archive; limited to OpenAI, Anthropic, and Meta/Llama models	Only one benchmark (MMLU); limited provider coverage; early-stage industry so trendline may shift
MIRI TGT ³	Jan 2025	Massive price dispersion across providers (up to ~10× for the same model); model size partially predicts price	Cross-provider pricing from Artificial Analysis; regression on parameter count vs. price	Author explicitly cautions against strong conclusions; narrow dataset; unclear how much cheapest providers are losing money
Introl Blog ⁴	Feb 2026	GPT-4-equivalent performance fell from ~$20 to ~$0.40/M tokens in ~3 years	Synthesis of Epoch AI, NVIDIA benchmarking, and market data	Secondary source; aggregates others' findings
iKangai ⁵	Aug 2025	Per-token costs fell ~1,000× in 3 years, but reasoning models consume 10–100× more tokens per task	Compared token generation counts across reasoning vs. non-reasoning models on identical queries	"Cost paradox" framing may overstate the issue for non-reasoning workloads
Don't Panic Labs ⁶	Dec 2025	Token prices dropping, but training costs doubling every ~8 months — current prices may be subsidized	Commentary on Epoch AI data + training cost trends	Opinion piece; limited original analysis

Trackers & Dashboards

Resource	URL	Notes
Price Per Token	pricepertoken.com	Daily-updated pricing across 300+ models; trend charts by lab
Artificial Analysis	artificialanalysis.ai	Provider-level pricing, speed, and quality comparisons
Epoch AI Data Insights	epoch.ai/data-insights	Benchmark-anchored cost trends with interactive charts

---

References

1. Epoch AI. "LLM inference prices have fallen rapidly but unequally across tasks." Epoch AI Data Insights, March 12, 2025. https://epoch.ai/data-insights/llm-inference-price-trends
2. a16z (Andreessen Horowitz). "Welcome to LLMflation." November 12, 2024. https://a16z.com/llmflation-llm-inference-cost/
3. MIRI TGT. "Observations About LLM Inference Pricing." Tech Gov Intelligence, January 2025. https://techgov.intelligence.org/blog/observations-about-llm-inference-pricing
4. Introl Blog. "Inference Unit Economics: The True Cost Per Million Tokens." February 9, 2026. https://introl.com/blog/inference-unit-economics-true-cost-per-million-tokens-guide
5. iKangai. "The LLM Cost Paradox: How 'Cheaper' AI Models Are Breaking Budgets." August 21, 2025. https://www.ikangai.com/the-llm-cost-paradox-how-cheaper-ai-models-are-breaking-budgets/
6. Don't Panic Labs. "The Price Per Token Is Dropping. Will It Stay That Way?" December 2, 2025. https://dontpaniclabs.com/blog/post/2025/12/02/the-price-per-token-is-dropping-will-it-stay-that-way/