LLM leaderboard: model release cadence Research cutoff: November 9, 2026 (America/New_York) RECOMMENDED CHARTS Use the labeled year-over-year release timeline to show exactly what shipped and when. Use the cumulative launch-day chart to compare release pace. Neither chart alone measures acceleration in model intelligence. To test intelligence gains per month, add a consistent independent benchmark series with stable evaluation methodology. WHAT THE DATASET COUNTS 61 plotted entries across Anthropic, OpenAI and Google's closed general-purpose model lines. Model sizes have separate entries. Anthropic Fable and Mythos versions share one entry because the company describes them as the same underlying model with different safeguards. GPT-5.1 and GPT-5.2 Instant/Thinking are grouped under their numbered launch rather than split into additional entries. These choices mean entry counts are not suitable as a strict cross-lab performance comparison. The pace chart instead counts distinct dates per company. Two or more eligible releases from the same company on the same date contribute one launch day. January 1 through November 9 is compared in both years. Matched-period launch days (2025 -> 2026): Anthropic: 5 -> 13 (2.6x) OpenAI: 5 -> 9 (1.8x) Google: 4 -> 8 (2.0x) These are descriptive counts of the included announcements, not statistical proof of increasing acceleration. Full-year 2025 launch days: Anthropic 6; OpenAI 7; Google 6. 2026 counts stop November 9; there is no extrapolation to year-end. DATES Use first official launch or preview announcements that describe external access, including restricted partner access. General-availability transitions, product-surface additions and redeployments are not counted again. This measures announced release cadence; announcements can precede user access during a phased rollout. Week position = 1 + elapsed days since January 1 / 7. This is not ISO week numbering. Points retain exact day positions on the axis. Vertical label offsets serve readability only. The gray area marks dates beyond the November 9 cutoff for 2026; late-2025 launches remain visible for context. IMPORTANT DATE RESOLUTIONS - GPT-5.6 Sol, Terra and Luna: June 26 restricted preview, rather than July 9 broad release. OpenAI's preview post names all three models and describes limited partner access. - Gemini 4 Argon: September 30 restricted rollout announcement, rather than October 13 general availability. - Gemini 4 Carbon: October 28 announcement, with partner access described as starting that week. This is not a general public release. - Claude Fable/Mythos 5.1: September 15 official newsroom announcement. Anthropic's covered-model help page lists August 31 as a designation date; that is not used as the launch announcement date. This is a consequential difference from a dataset built from designation or model snapshot dates. - Claude Fable/Mythos 5.2: November 9 announcement included. The earlier pasted summary's 12 Anthropic launch days is not reused: this reconstruction has 13 including today's announcement. - Gemini 2.0 Flash: excluded because the experimental model was introduced in December 2024; its 2025 GA transition is not a new first release. - Gemini 2.0 Flash-Lite: February 5 public preview used, rather than February 25 GA. - Gemini 2.5 Flash-Lite: June 17 preview used, rather than July 22 GA. EXCLUSIONS Coding-only variants (including Codex-specific releases), image/video/audio-only models, embeddings, cyber-only specialized variants, compute-only Pro modes, changed reasoning settings, price changes, fine-tuning changes, tools and agent products, routine snapshots under the same named model, rollout expansions and redeployments. Open-weight model lines such as Gemma and GPT-oss are outside this comparison. No fixed ranking is implied by labels such as flagship or smaller; product families do not form equivalent capability tiers across labs. SOURCES AND COVERAGE Each CSV row links an official announcement or official changelog. Research used the Anthropic Models newsroom index, individual Anthropic announcements, OpenAI announcements and API changelog, and Google announcements. This is a scoped reconstruction of named major releases, not an exhaustive list of every backend checkpoint or every access transition. The 56-event total in the pasted conversation is not used; its original CSV was not attached and its exact grouping rules cannot be reproduced. Primary source indexes: https://www.anthropic.com/news/models https://developers.openai.com/api/docs/changelog https://deepmind.google/models/gemini/ Source access note: Firecrawl returned insufficient-credit errors; research continued through web search and direct reads of official sources. Correction (November 9, 2026): Fable / Mythos 5.1 launched September 1, 2026, as confirmed by the official model overview and release notes. This corrects the prior September 15 date without changing the number of distinct launch days.