Verified developments ordered by when they actually happened — never by when EBIGRAPH discovered them.
ALPHAPrivate Alpha · Space Economy · NOW v2.19.05 · GRAPH 2.0 · VERIFIED INTELLIGENCE MAP
Sign in on EBIGRAPH to inspect current industry activity.
Private editorial workspace. Reviewer access required.
EBIGRAPH turns verified developments into decision context without blurring verified facts, interpretation and what remains unknown.
Verified developments that matter now — with EBIGRAPH interpretation, evidence and what to watch next.
Reality, with receipts. Every interpretation remains distinct from the underlying verified claim.
Promising developments with evidence, but not yet confirmed for NOW.
Items still requiring stronger evidence or resolution.
Super Intelligence (SI) & AGI: follow model releases through their sources. The sector name does not establish superintelligence or AGI capability. Public model-introduction reports are available separately; automated SI publication remains paused. Hourly discovery adds private review candidates; evidence still requires editorial review.
Review identitiesCompanies · Model developers, families and loaded private reviews.
Start with a company brief: what its loaded reviews report, why the changes may matter and what evidence is still missing. Expand a model family for the full review list. Active coverage is limited to 17 approved company groups. Priorities: frontier models, general agents, scientific discovery and enabling research. Inclusion is an editorial coverage choice, not a frontier-capability endorsement.
Loaded reviews are partial coverage. A listed family with no loaded records is a visible coverage gap, not evidence that the company has no models. Earlier out-of-scope records remain in the private archive; they are excluded from active coverage. All significance reviews remain unpublished.
Selected findings across companies: what changed, why it may matter and the evidence still needed. Saved assessments can remain held; none of these cards approves a breakthrough.
Company names identify model developers; a product may serve models from other companies. Cursor appears under Anysphere with Composer and its coding product. Cursor/Anysphere is grouped editorially with xAI; this grouping does not assert ownership. Developer identity and model lineage remain separately visible. Official links support family identity, not breakthrough significance.
The queue includes a preliminary Astra evidence assessment, held for further review. No breakthroughs are approved in this view. The other records are review materials, not labelled breakthroughs. A corroborated announcement establishes that something happened; it does not establish the size or importance of an advance.
A defined task becomes feasible, or a meaningful threshold is crossed. Name the prior limit, comparison baseline and evaluation conditions.
A substantial reduction in cost for comparable successful work. Token price alone is insufficient; include quality, reasoning usage and actual task cost.
A capability becomes materially more accessible. Distinguish announced plans, restricted pilots, hosted APIs and downloadable weights; inspect license conditions.
A demonstrated improvement in a real workflow, with scale, limitations and repeatability explained. A benchmark rank alone is insufficient.
Approval requires claim-matched evidence, a dated baseline, independently examined results, limitations and an editorial explanation of significance. These are proposed criteria, not an automated scoring rule.
Counts reflect your search within the loaded private records, including the Astra editorial note.
Loading saved review status…
Inspect the linked evidence for evaluation methods, proof artifacts and source-version discrepancies. Formal proof checking and independent replication are separate questions.
Save to the private database to access reviews across devices. Browser drafts remain available as a local backup. Reviews stay unpublished.
Explore a private claim and its original source and receipts. Select the claim to trace its revision and source fingerprint. Select a receipt to inspect its stored evidence, claim-match and independence review status. Publication remains paused.
Private reviewed claims, not a public verified chronology. Dates belong to the reviewed claims; held records may still need corroboration. Source variants can describe the same event.
Bundled benchmark snapshot: Oct 5, 2026 · prices checked Oct 5.
Explore published measurements. These are benchmark records, separate from EBIGRAPH event verification.
Capability: Epoch AI, Epoch Capabilities Index (ECI), CC BY 4.0. ECI combines multiple benchmark sources, including developer-reported results; it is not a guarantee of real-world effectiveness. Scores and uncertainty ranges come from the dated source download shown above. ECI is not a percentage and has no fixed maximum.
Price: official provider documentation checked Oct 5. USD per million tokens; standard uncached API rates, DeepSeek peak rates. GPT-6 Astra rates apply to requests with at most 272K input tokens. No batch, caching, taxes or tool fees. The token mix is illustrative, not measured task cost. Different tokenizers and reasoning usage affect actual bills. Only exact model matches with checked prices are plotted; other prices remain unavailable. Benchmark downloads refresh daily. Prices remain manually checked and dated separately.
Epoch methodology ↗ · Data & attribution ↗ · CC BY 4.0 ↗Research capability, the cost of useful work and practical deployment. These selected findings come from held assessments; they do not establish approved breakthroughs or AGI.
Briefing prepared October 9, 2026 · Evidence snapshots examined October 5–8. The briefing covers all its selections; the company, search and status filters below apply to the detailed findings.
Follow-up priorities: proof and novelty examination for Astra; matched successful-work economics for Sol and GLM Flash; reproducible evaluator research for Amazon.
What changed, why it may matter and the evidence still missing. Each finding keeps its sources and review status.
Follow the same questions across companies: reasoning, agents, scientific discovery, multimodal capabilities and the cost of useful work. Reported measurements and developer claims remain attached to their individual reviews; this view does not establish independent validation or a cross-company ranking.
Search covers the records loaded below.
Compare reviewed claims and evidence. This is not a performance ranking; records may concern different release stages.
Related claims found across sources. Matching does not establish independent corroboration; all results remain private.
Private records with announcement review or saved significance drafts; extraction and verification may still be pending. Multiple language versions can describe the same event; they are not independent sources.
Coverage preparation includes US and Chinese model developers. Preview records represent reviewed pilot claims, not a complete industry feed.
Loading graph…