Independent AI systems researchOperated by NeuralArc
AccessAllGPTRESEARCH
Menu
Evidence index

Inspect the source before the verdict.

This index exposes the material behind every decision guide: what we reviewed, when we retrieved it, the claim it supports and where the evidence stops.

181decision guides
790source records
782scoped claim notes
How to read this index

A citation is not a transferable result.

Reviewed identifies the part of the source used. Supports states the bounded proposition we draw from it. Retrieved records when time-sensitive material was checked.

Vendor documentation establishes vendor claims and product behavior as documented, not independent performance. Papers establish results only for their stated methods and tested scope. AccessAllGPT recommendations remain labeled analysis.

Read the full methodology →
Source ledger

Evidence by decision guide

Ordered by the latest publication or material update date. Follow a source to inspect the original material.

Models

Gemini 3.8 Live Avatar Is Generally Available—Here Are the API Price and Limits

AccessAllGPT reviewed five first-party Google sources on September 27, 2026: the September 24 launch article, September 25 Google Cloud general-availability announcement, Live API documentation last updated September 25, live Agent Platform pricing table, and September 15 Gemini 3.8 Audio model card. We checked announcement date, availability, exact model ID, endpoints, modalities, protocol, pricing, context and output limits, session duration, language statements, avatar controls, watermarking and known limitations. Launch and performance statements are labeled as Google claims; API fields and prices are documented service facts; rollout guidance is AccessAllGPT analysis. We did not have a Google Cloud credential or enterprise allowlist, so we did not call the API, create an avatar, inspect a response receipt, test tool execution, verify SynthID detection or measure latency, quality, language performance, session duration or cost.

5 sources · 5 retrieval dates · 5 scoped claim notes

  1. Introducing Gemini 3.8 Live with Live Avatar

    Google · Retrieved

    Reviewed: September 24, 2026 announcement date; Gemini Enterprise availability; multimodal conversation; asynchronous tool execution; 97-language claim; custom-avatar allowlist; SynthID; API link

    Supports: Google announced Gemini 3.8 Live with Live Avatar on September 24, says it is available in Gemini Enterprise, and describes near-real-time audio and video, background tool calls, 97-language transitions, preset and allowlisted custom avatars, and SynthID watermarking. These capability and quality statements are vendor claims.

  2. Power your agents: Gemini 3.8 Live with Live Avatar is now generally available

    Google Cloud · Retrieved

    Reviewed: September 25, 2026 publication date; general availability; enterprise production positioning; features; US and EU endpoints; provisioned throughput; custom-avatar controls; SynthID; demos

    Supports: Google Cloud calls Live Avatar generally available and ready for enterprise production in Gemini Enterprise. It lists web, mobile and kiosk use; asynchronous tools; live visual input; US and EU endpoints; provisioned throughput; curated avatars; allowlisted custom avatars; and SynthID. Gemini 3.8 Live Extended Thinking remains private preview.

  3. Gemini Live API overview

    Google Cloud Documentation · Retrieved

    Reviewed: Overview; key features; technical specifications; supported models; gemini-3.8-live availability and feature row; protocols and modalities; page update date

    Supports: The documentation lists model ID gemini-3.8-live as generally available with native audio, transcriptions, voice activity detection, affective dialogue, proactive audio, tool use and Live Avatar. It documents a stateful WebSocket protocol, text/audio/image/video inputs, audio/text/avatar-video outputs, and 24 supported conversational languages. The page was last updated September 25, 2026 UTC.

  4. Generative AI on Agent Platform pricing

    Google Cloud · Retrieved

    Reviewed: Gemini 3 standard-model pricing table; Gemini 3.8 Live API rows; modality-specific input and output prices; region; cache columns; billable-response rule

    Supports: Google lists non-global Gemini 3.8 Live API prices per one million tokens: $0.75 text input, $1 image/video input, $3 audio input, $4.50 text output including response and reasoning, $12 audio output, and $1 avatar-video output. The table shows no cached-input rates and says only requests returning HTTP 200 are charged for input or output.

  5. Gemini 3.8 Audio model card

    Google DeepMind · Retrieved

    Reviewed: Publication date; model dependencies; inputs and outputs; context and output limits; distribution; intended use; known limitations; evaluation and Frontier Safety summaries

    Supports: The model card says Gemini 3.8 Audio is based on Gemini 3 Pro. For Live it lists audio, image, video and text input with up to 128K context and 64K output; with Live Avatar it lists audio, video and text output with a 24K output limit. It names hallucination, occasional slowness or timeouts, a January 2025 knowledge cutoff and only a few minutes of continuous Live Avatar interaction as limitations.

Papers to production

Claude Computes a Nine-Loop Physics Result—With Validation and a $1,000–$2,000 Cost Estimate

AccessAllGPT reviewed three public sources on September 27, 2026: Anthropic’s September 25 guest post and validation addendum, the public Cosmic9 result repository, and Matt von Hippel’s August 7 pre-commitment challenge. We checked the target, model and harness identity, prompts, human involvement, two computational routes, resource and end-user cost estimates, validation statements, concurrent result, artifact availability, disclosed conflicts, and limits on generalization. Anthropic’s narrative and prices are labeled as reported claims; public files and documentation are artifact facts; the original challenge establishes that the target was named before the attempt; deployment guidance is AccessAllGPT analysis. We did not rerun the computation, download the hundreds-of-megabytes archives, reconstruct coefficients, inspect private Claude traces, or peer-review the physics.

3 sources · 3 retrieval dates · 3 scoped claim notes

  1. Yes, Claude can do Nine Loops

    Anthropic (guest post by Matt von Hippel, with addendum by Lance Dixon) · Retrieved

    Reviewed: September 25, 2026 publication date; challenge; nine-loop computation; prompts; methods; compute and user-cost estimates; concurrent human result; interpretation; Lance Dixon validation addendum; downloadable materials; disclosure

    Supports: Anthropic published the account on September 25. It says Fable 5.1 in Claude Science computed a six-particle planar N=4 super-Yang-Mills amplitude at nine loops by two routes, with limited human direction. The post estimates either route would cost an end user $1,000–$2,000 and says the direct bootstrap used about $100 of 96-CPU compute for one week. Independent physicist Lance Dixon describes how he validated the result. The post discloses that Anthropic compensated the guest author and gave Dixon Claude usage credits.

  2. Cosmic9: Cosmically Normalized Six-Point Amplitudes at Nine Loops

    Siddharth Mishra-Sharma · Retrieved

    Reviewed: Status; method summary; two representations; rational and modular certification; validation notes; conventions; manifests and checksums; downloadable amplitude, symbol-space and function artifacts; Zenodo archive reference

    Supports: The result page publishes machine-readable files for the nine-loop amplitude, explains the form-factor and direct-bootstrap routes, says the two representations agree on every compared coefficient, identifies assumptions and untested items, and provides a SHA-256 manifest plus validation records. It also states that the computation programs are not distributed and hosts files over 100 MB separately on Zenodo.

  3. It Only Counts When AI Gets to My Field

    4 gravitons (Matt von Hippel) · Retrieved

    Reviewed: August 7, 2026 challenge; definition of a meaningful result; named N=4 super-Yang-Mills nine-loop target; computational-difficulty rationale; expected academic-compute constraint; author biography and conflict context

    Supports: Before Anthropic attempted the work, von Hippel publicly named the six-particle N=4 super-Yang-Mills amplitude at nine loops as one result that would count as meaningful AI progress in his former field. The post framed the target as computationally difficult but solvable in principle with known methods, and required resources available to an academic rather than extraordinary compute.

Agents

AWS Publishes a Dual-Monitoring Blueprint for Production AI Agents

AccessAllGPT reviewed the first-party AWS architecture post published September 11, 2026, the live AgentCore Evaluations developer page and the live AgentCore pricing page on September 14, 2026; all three returned HTTP 200. We recorded the AWS article’s canonical link to the sample in the aws-samples GitHub organization, but direct GitHub retrieval failed after retries because the publishing environment could not establish a connection. We checked the announcement date, architecture, evaluator catalog, sampling range, telemetry path, IAM example, stated caveats, sample identity and evaluation rates. We did not deploy the sample, create an AWS account, run an evaluator, submit production traces or test AWS DevOps Agent. Service behavior and performance statements are AWS claims; prices are the public rates retrieved on September 14; recommendations are AccessAllGPT analysis.

4 sources · 4 retrieval dates · 4 scoped claim notes

  1. Monitoring production agent lifecycle with AWS DevOps Agent and AgentCore Evaluations

    AWS Artificial Intelligence Blog · Retrieved

    Reviewed: September 11, 2026 publication metadata; architecture; built-in evaluators; online and on-demand evaluation; responsible AI safeguards; AWS DevOps Agent investigation; adoption notes; open-source sample

    Supports: AWS published a dual-monitoring reference architecture for multi-agent systems using AgentCore Evaluations for sampled quality scoring and AWS DevOps Agent for infrastructure investigation. The post documents a four-agent airline example, 16 built-in evaluators, 0.01–100 percent online sampling, OpenTelemetry and CloudWatch integration, an IAM failure demonstration, caveats and source code.

  2. Evaluate agent performance with Amazon Bedrock AgentCore Evaluations

    Amazon Bedrock AgentCore Developer Guide · Retrieved

    Reviewed: Service overview; supported instrumentation; evaluator resource and IAM model; quotas; online, on-demand, batch and dataset evaluation topic index

    Supports: AWS documents AgentCore Evaluations as automated assessment before and after deployment, using OpenTelemetry or OpenInference traces and LLM-as-a-Judge techniques for built-in and custom evaluators. The page also documents private custom evaluator resources, IAM policies and regional configuration quotas.

  3. Amazon Bedrock AgentCore pricing

    Amazon Web Services · Retrieved

    Reviewed: Evaluations consumption model; price table for built-in, custom and batch evaluations; CloudWatch charges; sampling cost controls; optimization preview conditions

    Supports: AWS lists built-in evaluation at $0.0024 per 1,000 input tokens and $0.012 per 1,000 output tokens, custom evaluation at $1.50 per 1,000 evaluations plus model usage, and batch evaluation at $0.0018 per 1,000 input tokens and $0.009 per 1,000 output tokens. CloudWatch telemetry charges are separate.

  4. Dual monitoring system sample

    AWS Samples on GitHub · Retrieved

    Reviewed: Repository route and sample identity as linked from the September 11 AWS article; direct GitHub retrieval was attempted but blocked by network exhaustion

    Supports: The AWS article links a sample implementation in the aws-samples organization for the dual-monitoring system it describes. AccessAllGPT recorded that first-party link but could not retrieve, deploy, execute or security-review the repository.

Models

Google Launches the Gemini App Globally on Windows 10 and 11

AccessAllGPT reviewed Google’s September 10, 2026 first-party launch post and the live Gemini Apps Privacy Hub on September 13, 2026. The announcement returned HTTP 200 and was checked for date, operating-system support, global availability, shortcut, product surfaces and its subscription footnote. The privacy page returned HTTP 200 and was reviewed for activity, retention, Connected Apps and Spark data boundaries. The linked gemini.google/desktop installer page was unreachable from our publishing environment after repeated connection attempts, so we did not inspect the installer, verify binary signatures, install the app, authenticate an account or test performance and features. Product facts and capability statements below are Google claims; operational recommendations are AccessAllGPT analysis.

2 sources · 2 retrieval dates · 2 scoped claim notes

  1. The Gemini app is now available for Windows

    Google · Retrieved

    Reviewed: September 10, 2026 publication date; launch summary; keyboard shortcut; workspace, Connected Apps and media features; global Windows availability; subscription footnote

    Supports: Google announced the Gemini app for Windows on September 10, 2026 and says it is available globally for Windows 10 and 11. The post documents Alt + Space access, a dedicated Gemini workspace, Gmail and Drive connections, image and video creation, and an AI-subscription requirement for Gemini Spark and Gemini Omni with availability varying and an 18+ condition.

  2. Gemini Apps Privacy Hub

    Google Gemini Apps Help · Retrieved

    Reviewed: Gemini Apps Privacy Notice; data collected and used; Keep Activity; retention; Connected Apps; Gemini Spark remote browser and remote computer; work and school account distinction

    Supports: Google says Gemini can process prompts, uploads, generated content, Connected Apps data and device or browser context. Keep Activity controls whether future chats appear in activity and are used to improve AI models, subject to stated exceptions; Spark can save task files and share necessary information with services and third parties. Work and school accounts may follow different terms.

Models

AWS Publishes an OpenAI-on-Bedrock Benchmark—and Luna Wins Its Cost-per-Outcome Samples

AccessAllGPT reviewed AWS’s September 11, 2026 report, its July 30 pricing announcement, the live Bedrock pricing page and the public benchmark repository on September 13, 2026. We pinned repository commit d0e60ba1050be7cf880a4ca7b206c1442088d3aa and inspected its README plus selected checked-in AIME, GPQA, DeepSearchQA and GDPval JSON artifacts. We confirmed the named backends, model IDs, sample counts, reasoning settings, grading disclosures, repricing factors and selected headline values from those files. We did not execute the paid model benchmarks, reproduce web searches, inspect every artifact or independently validate answer quality. Results are AWS/OpenAI-authored vendor evidence; interpretations and adoption gates are AccessAllGPT analysis.

4 sources · 4 retrieval dates · 4 scoped claim notes

  1. Beyond the price per token: Choosing the right OpenAI model on Amazon Bedrock for your workload

    Amazon Web Services · Retrieved

    Reviewed: September 11, 2026 publication date; solution overview; cost per correct answer; agent trajectory cost; GDPval; decision framework; latency; reproduction commands; conclusion

    Supports: AWS published results for GPT-5.6 Luna, Terra and Sol on Amazon Bedrock against GPT-5.4 mini and nano on the OpenAI API. The post states sample sizes, model configurations, reported quality and cost outcomes, pricing assumptions, latency scope, limitations and reproduction commands.

  2. openai-on-aws/benchmarks-openai at commit d0e60ba

    AWS and OpenAI benchmark repository on GitHub · Retrieved

    Reviewed: README; suite descriptions; backend and model identifiers; quality methodology; checked-in AIME, GPQA, DeepSearchQA and GDPval result JSON; judge caveats; repricing records; license

    Supports: The pinned repository exposes the harness and timestamped result artifacts behind the AWS post. AccessAllGPT checked the recorded sample sizes, model IDs, reasoning settings, token and cost fields, grading thresholds, judge disclosure and selected headline values without making paid model calls.

  3. Amazon Bedrock announces up to 80% lower prices for OpenAI GPT-5.6 models

    Amazon Web Services · Retrieved

    Reviewed: July 30, 2026 date; Luna and Terra reductions; Sol statement; effective date; automatic application; supported US regions; Responses API access

    Supports: AWS says on-demand Bedrock pricing fell 80% for GPT-5.6 Luna and 20% for GPT-5.6 Terra effective July 30, 2026, while Sol pricing was unchanged. It lists Luna and Terra access in three US regions through the OpenAI Responses API on bedrock-mantle.

  4. Amazon Bedrock pricing

    Amazon Web Services · Retrieved

    Reviewed: OpenAI provider section; frontier-model selector; GPT-5.6 Luna, Terra and Sol entries; pricing-tier labels; region-dependent table structure

    Supports: The live pricing page listed GPT-5.6 Luna, Terra and Sol in its OpenAI model selector on retrieval. Its initial server-rendered view exposed a different default model table, so AccessAllGPT did not treat that view as independent confirmation of the exact Luna or Terra rates recorded by the benchmark.

Agents

Duplicate JSON Keys Can Split an AI Agent’s Approval From Execution

AccessAllGPT wrote and ran scripts/audit-agent-json-duplicate-keys.mjs on 2026-09-12 using Node v24.10.0 on Darwin arm64, Python 3.9.6 and Ruby 2.6.10. Five fixed JSON tool-call fixtures covered one unique-key control and four duplicate-member cases: top-level tool, nested path, a Unicode-escaped equivalent of path and a nested approval flag. A small event-preserving scanner decoded member names and rejected duplicates before object construction. We compared each runtime’s parsed value structurally, measured a deliberately unsafe first-match approval preview against the Node execution object, and checked what a JSON.parse reviver could observe. The deterministic artifact is artifacts/agent-json-duplicate-keys-audit.json; its embedded pre-self-reference SHA-256 is 702396229933fe5ed3d6abaf5fef13ed01df1a1eadf8b5dbf3516efe22181127 and the final file SHA-256 is ab55262b645882d2d2698abef3f8def51b064c225ecf316a40c3d147d8fe1570. Standards language, runtime documentation, vendor guidance, local observations and AccessAllGPT recommendations are labeled separately.

4 sources · 4 retrieval dates · 4 scoped claim notes

  1. RFC 8259: The JavaScript Object Notation (JSON) Data Interchange Format

    RFC Editor / IETF · Retrieved

    Reviewed: Section 4 Objects, including member-name uniqueness and interoperability behavior

    Supports: RFC 8259 says object names SHOULD be unique, objects with unique names are interoperable, and behavior is unpredictable when names are not unique; it notes that many implementations report only the last pair while others error or preserve duplicates.

  2. json — JSON encoder and decoder: Standard Compliance and Interoperability

    Python Software Foundation · Retrieved

    Reviewed: Repeated Names Within an Object and the default decoder’s documented behavior

    Supports: Python documents that its json module accepts repeated names by default and uses only the value of the last name-value pair. This independently documents one runtime behavior; it does not define a safe agent protocol.

  3. JSON.parse()

    MDN Web Docs · Retrieved

    Reviewed: Syntax, return value, reviver traversal and examples for parsing JSON text

    Supports: MDN documents JSON.parse conversion and post-parse reviver traversal. The local harness, rather than MDN, establishes that the tested reviver received one decoded path key after duplicate collapse.

  4. Safety in building agents

    OpenAI Developer Documentation · Retrieved

    Reviewed: Structured outputs, tool approvals, untrusted data flow, guardrails and evaluation guidance

    Supports: OpenAI’s vendor-authored guidance recommends structured outputs and tool approvals among layered agent controls. It does not specify duplicate-member handling and is not evidence that those controls are safe when approval and execution consume different representations.

Agents

Funes Gives AI Agents Memory—Not a Trusted Instruction Channel

AccessAllGPT reviewed the September 3 Funes launch, v1.3.0 release, documentation, security policy, source at commit 72c108157ffd5721a1cbb68e3394cfa5eeaaac7f and the published handoff-vs-recall benchmark on 2026-09-12. We downloaded the versioned release asset rather than piping the installer, verified the macOS arm64 binary as funes 1.3.0 with SHA-256 3d9c8aec11cbd604a94dde72753d0683ec289cf0dd6748eb03a8355d2c258826, and ran it under isolated HOME and XDG directories. A synthetic four-turn Codex trace contained one useful queue decision and one explicit instruction-like injection but no real credentials. Funes indexed one session into four chunks in 0.17 seconds; scan and get returned both passages verbatim with provenance. Indexing warned that TruffleHog was absent and continued locally. We did not bind an agent, publish a memory, execute recalled text, download the 1.11 GB reranker, or claim a semantic-recall benchmark. Source hashes and run results are preserved in artifacts/funes-1-3-memory-trust-boundary-audit.json and validated by scripts/audit-funes-memory-boundary.mjs. Vendor product claims, author-run benchmark evidence, independent threat evidence, local observations and AccessAllGPT guidance are labeled separately.

9 sources · 9 retrieval dates · 9 scoped claim notes

  1. Give Your Coding Agents a Memory You Own

    Hugging Face · Retrieved

    Reviewed: Publication date; installation; indexing and recall pipeline; local and shared memory; secret handling; benchmark summary; supported agents

    Supports: Hugging Face launched Funes as a durable memory layer for Claude Code, Codex, pi and Hermes. The article describes local embedding and reranking, optional private Hub datasets, raw-passage provenance and an author-run two-task benchmark.

  2. funes v1.3.0 release

    Hugging Face on GitHub · Retrieved

    Reviewed: Release timestamp; binary assets; VERSION and SHA256SUMS assets; platform coverage

    Supports: GitHub records v1.3.0 on September 1, 2026 and publishes three platform binaries plus release checksums. This fixes the tested product version and artifact boundary.

  3. funes source tree at commit 72c1081

    Hugging Face on GitHub · Retrieved

    Reviewed: Cargo manifest; indexing, recall, scan, push and secret-scanner code; tests; Apache-2.0 license

    Supports: The inspected source pins the Lance storage dependency and local embedding/reranking models, implements hybrid retrieval and recency weighting, and separates best-effort index-time redaction from a fail-closed publish gate.

  4. Funes security policy

    Hugging Face on GitHub · Retrieved

    Reviewed: Local processing; remote memory; secret scanning; incident response; untrusted memories; token scope

    Supports: The project explicitly says third-party memory is untrusted input capable of prompt injection, recommends fine-grained tokens, and warns that deleting later commits does not erase an exposed secret from repository history.

  5. Funes indexing and recall documentation

    Hugging Face on GitHub · Retrieved

    Reviewed: Supported trace roots; Parquet schema; incremental IDs; indexing tiers; data-URI handling; model pin; storage pipeline

    Supports: The documented index keeps raw text, thinking, tool calls and tool results in tiers, uses BAAI/bge-small-en-v1.5 embeddings, and derives deterministic chunk IDs from session, turn, block and split coordinates.

  6. Funes publishing and sharing documentation

    Hugging Face on GitHub · Retrieved

    Reviewed: First-push guard; TruffleHog behavior; block reconstruction; hold-back and scrub; session selection; append-only remote behavior

    Supports: The push boundary requires TruffleHog and reconstructs complete blocks before scanning. It can withhold secret-bearing rows, but selection is pre-publication only and no command retracts an already-published session.

  7. Funes handoff-vs-recall benchmark results

    dacorvo on Hugging Face Datasets · Retrieved

    Reviewed: Aggregate table; two task definitions; arm-level repetitions; token accounting; failures; reconstructed receipt disclosure; contamination controls

    Supports: The author-run benchmark reports 30 runs over two project-specific tasks. Recall passed all six of its runs and used fewer weighted tokens per successful task than the compared successful channels; one handoff preparation figure is reconstructed rather than read from a surviving receipt.

  8. LLM01:2025 Prompt Injection

    OWASP GenAI Security Project · Retrieved

    Reviewed: Description; direct and indirect injection; impacts; prevention and mitigation; RAG attack scenario

    Supports: OWASP describes indirect prompt injection through external files and websites, says RAG and fine-tuning do not fully mitigate it, and recommends least privilege, human approval, segregating external content and adversarial testing.

  9. AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents

    arXiv; authors from ETH Zurich and Google DeepMind · Retrieved

    Reviewed: Abstract; evaluation framing; tool-using agent threat model; benchmark scope and version history

    Supports: Independent researchers frame tool-returned untrusted data as an agent prompt-injection surface and introduce a benchmark for measuring attacks and defenses. It supports the threat model, not a finding about Funes specifically.

Models

Lyria 3.5 Is an $0.08 Full-Song API—but Its Java Samples Still Name Lyria 3

AccessAllGPT reviewed Google’s September 4 launch, the live Lyria family page, Gemini API music-generation guide, pricing table, API terms, Lyria prompt guide and SynthID page, retrieved 2026-09-05. We wrote and ran scripts/audit-lyria35-contract.mjs on Node v24.10.0, Darwin arm64. The audit fetched all seven public pages, asserted the launch date, model IDs, 44.1 kHz output, price rows, ten-image limit, single-turn boundary, paid-service data terms and the repeated legacy Java identifier, and wrote response status, byte counts and SHA-256 hashes to artifacts/lyria35-contract-audit.json. No Gemini API credential was present, so we did not generate or listen to audio, measure duration or latency, test prompt adherence, inspect a binary watermark, or validate billing. Google supplies the only currently located Lyria 3.5 capability evidence; no independent model-specific evaluation was available at publication.

7 sources · 7 retrieval dates · 7 scoped claim notes

  1. Create your best tracks yet with Lyria 3.5 in Gemini

    Google · Retrieved

    Reviewed: September 4 publication date; launch summary; Gemini app and API availability; track controls; global and product availability

    Supports: Google announced Lyria 3.5 on September 4, 2026 and says it is available in the Gemini app, Gemini API, Google AI Studio, Flow Music and Google Vids. Expressive-vocal, arrangement and fidelity statements are vendor claims.

  2. Lyria 3.5 model family page

    Google DeepMind · Retrieved

    Reviewed: Lyria 3.5 introduction; model family; capabilities; prompt examples; artist partnerships; safety; limitations; product access surfaces

    Supports: The family page says Lyria 3.5 can generate tracks up to three minutes, accepts image prompts, supports lyrics and multiple languages, exports high-fidelity audio and embeds SynthID. It provides demonstrations, not a controlled evaluation.

  3. Generate music with Lyria 3.5

    Google AI for Developers · Retrieved

    Reviewed: Model table; Interactions API; full-song generation; response formats; response schema; image inputs; custom lyrics; timestamps; languages; limitations; code examples

    Supports: The live API contract names lyria-3.5 for full songs and lyria-3-clip-preview for fixed 30-second clips, with 44.1 kHz stereo MP3 output, optional WAV for 3.5, up to ten images, single-turn generation, nondeterminism, safety filters and SynthID. Its Java examples repeatedly name the older lyria-3-generate-001 identifier.

  4. Gemini Developer API pricing

    Google AI for Developers · Retrieved

    Reviewed: Lyria 3.5 and Lyria 3 price rows; free-tier availability; paid per-request rates; product-improvement data-use labels

    Supports: The September 5 pricing page lists no free API tier for Lyria music generation, $0.08 per Lyria 3.5 full song, $0.04 per Lyria 3 Clip Preview request and $0.08 per legacy Lyria 3 Pro Preview full song. The table labels paid-tier content as not used to improve products.

  5. Gemini API Additional Terms of Service

    Google AI for Developers · Retrieved

    Reviewed: Age and region requirements; API-client restrictions; generated content; unpaid and paid service data use; logging; payment and disclaimers

    Supports: The terms require API users and API-client audiences to be 18 or older, limit use to professional or business purposes, distinguish unpaid from paid data use, disclaim exclusivity in generated content and leave legal responsibility with the developer.

  6. How to create effective prompts with Lyria

    Google DeepMind · Retrieved

    Reviewed: Simple, image and structured prompts; genre; tempo; instruments; dynamics; vocals; supplied lyrics; experimentation guidance

    Supports: The prompt guide recommends specifying genre, tempo, instrumentation, dynamics, vocal profile and lyrics, and presents image-to-music as mood interpretation. It supplies a useful test-factor inventory but no adherence rates.

  7. SynthID: Identifying AI-generated content with new watermarking technology

    Google DeepMind · Retrieved

    Reviewed: Audio watermarking overview; detection positioning; robustness claims; deployment scope; limitations and responsible-AI framing

    Supports: Google describes SynthID as an imperceptible watermark and detector for AI-generated audio and other media. This is first-party technology documentation, not an independent false-positive, false-negative or transformation-robustness evaluation of Lyria 3.5 outputs.

Benchmarks

RAG Chunk Overlap: 25% Recovered Split Evidence; 75% Tripled the Index

AccessAllGPT wrote and ran scripts/audit-rag-overlap-crowding.mjs with Node v24.10.0 on Darwin arm64 on 2026-09-05. The deterministic synthetic corpus contains 24 documents of 240 words and 12 queries. Each target places a required cause token at word 77 and answer token at word 80 around an 80-word chunk boundary; a paired distractor shares topic and cause terms but not the answer. We compared 0%, 25%, 50% and 75% overlap with fixed top-k=4 and an explicitly simple query-term-frequency retriever. The run emitted artifacts/rag-overlap-crowding-audit.json; its embedded pre-self-reference SHA-256 is d7dbda9bb754342061a14ab016618debd2d8a0064f4e103d74155d00f7bf73f5, and the final JSON file SHA-256 is 2386a468336133f2473d653fbb48ff9da7b0fb06348fed2632b27dbc7b4883f5. We also relied on our 2026-08-06 review of the original RAG paper and OpenAI’s vendor implementation guidance. Academic evidence, vendor guidance, local observations and AccessAllGPT recommendations are labeled separately.

2 sources · 2 retrieval dates · 2 scoped claim notes

  1. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (version 4)

    arXiv (authors from Facebook AI Research, University College London and New York University) · Retrieved

    Reviewed: Abstract, introduction, methods, experiments, results, discussion and limitations of the paper’s experimental scope

    Supports: The authors define and evaluate a specific RAG architecture combining parametric sequence-to-sequence memory with retrieved non-parametric Wikipedia passages. The paper establishes retrieval as part of the generation system in its tested scope; it does not prescribe a universal chunk size or overlap.

  2. Optimizing LLM Accuracy

    OpenAI Developer Documentation · Retrieved

    Reviewed: LLM optimization context, prompt engineering, retrieval-augmented generation, fine-tuning and production accuracy sections

    Supports: OpenAI’s vendor-authored guidance distinguishes context optimization for missing, stale or proprietary knowledge from model optimization for inconsistent behavior and recommends an evaluate–hypothesize–change–evaluate cycle. It is implementation guidance, not independent proof for an overlap setting.

Models

Claude Fable 5.1 Cut Cache Reads 75%—but It Is Not a Drop-In Agent Upgrade

AccessAllGPT reviewed Anthropic’s launch page, live model overview, what-is-new page, migration guide, pricing table and 212-page system card, plus AWS availability evidence and Harbor’s public Terminal-Bench-Science surface, all retrieved 2026-09-04. We downloaded the system-card PDF, recorded SHA-256 b0d59edc7a60eef32a879c13d713cce60c3fefd7e6b5183afdc8b835af3c8c39, and extracted 212 pages with pypdf 6.17.0 under Python 3.11.15. We wrote and ran scripts/audit-fable51-contract.mjs under Node to fetch the live docs, assert the model ID, context/output limits, price rows and migration warnings, and calculate transparent cache-workload scenarios. No Anthropic API credential or Fable 5.1 entitlement was available, so we did not claim model-output, latency, tool-use or safety reproduction.

8 sources · 8 retrieval dates · 8 scoped claim notes

  1. Claude Fable 5.1 and Claude Mythos 5.1

    Anthropic · Retrieved

    Reviewed: Introduction; performance; scientific research; safety, security and alignment; Mythos; cost and availability; benchmark notes

    Supports: Anthropic introduces Fable 5.1 as generally available and Mythos 5.1 as the same base model behind a restricted safeguards/access layer. It states unchanged input/output rates, lower cache-read pricing, phased Enterprise Frontier Safeguards, effort defaults, benchmark results and safety claims. These are vendor claims.

  2. Claude Fable 5.1 model overview

    Anthropic Claude Platform documentation · Retrieved

    Reviewed: Model ID; specifications; overview; breaking changes; comparison; availability; launch resources

    Supports: The live product contract names claude-fable-5-1, a 1M-token context window, 128K maximum output, $10/MTok base input and $50/MTok output. Anthropic recommends Opus 5 first for most workloads and Fable only when demanding long-horizon evaluations justify it.

  3. What's new in Claude Fable 5.1

    Anthropic Claude Platform documentation · Retrieved

    Reviewed: Models and availability; context; thinking and effort; breaking changes; progress updates; pricing; content provenance

    Supports: The release changes forced tool use, thinking-block compatibility and conversation-edit behavior. It adds beta per-message effort, turn-scoped system messages, progress updates, content provenance and a lower cache-hit price.

  4. Migrating to Claude Fable 5.1 and Claude Mythos 5.1

    Anthropic Claude Platform documentation · Retrieved

    Reviewed: Baseline settings; platform availability; migration paths; breaking changes; checklists; code examples

    Supports: The migration guide documents 400 responses for unsupported thinking configurations and assistant prefill, the forced-tool incompatibility, model-bound thinking blocks, invalidation after editing earlier turns, and platform-specific migration work.

  5. Pricing

    Anthropic Claude Platform documentation · Retrieved

    Reviewed: Model pricing table; cache write and hit columns; Fable 5.1, Fable 5 and Opus 5 rows

    Supports: At retrieval, Fable 5.1 costs $10/MTok base input, $12.50 for five-minute cache writes, $20 for one-hour cache writes, $0.25 for cache hits/refreshes and $50 for output. Fable 5 cache hits cost $1; Opus 5 is $5/$0.50/$25 for base input/cache hit/output.

  6. Claude Fable 5.1 & Claude Mythos 5.1 System Card

    Anthropic · Retrieved

    Reviewed: Executive summary; autonomy and alignment; cyber and bio; safeguards; behavioral audits; capability evaluation methods and limitations

    Supports: The 212-page card reports model and safeguard evaluations, including a vendor-run capability table, methodology details, external METR testing, uncertainty, and disclosed regressions. It says Mythos 5.1 slightly regressed on overall misaligned behavior versus Opus 5 while improving versus Mythos 5 and Sonnet 5.

  7. Claude Fable 5.1, Anthropic's new frontier model is now available on AWS

    Amazon Web Services · Retrieved

    Reviewed: September 1 posting date; availability statement; Amazon Bedrock and Claude Platform on AWS access paths

    Supports: AWS independently corroborates general availability on two AWS surfaces as of September 1, 2026. It does not independently validate Anthropic capability or safety claims.

  8. Terminal-Bench-Science 0.1 dataset and leaderboard

    Harbor Framework · Retrieved

    Reviewed: Dataset identity; research-workflow scope; revision 10 / 0.1.0; public leaderboard surface

    Supports: Harbor exposes the public benchmark as a research-workflow dataset spanning scientific disciplines. The browser-accessible leaderboard shell did not expose a Fable 5.1 row at retrieval, so it cannot independently confirm Anthropic’s new score.