OpenAI Assistants API was sunset on August 26, 2026
OpenAI’s Assistants API reaches its shutdown deadline on August 26, 2026. OpenAI recommends Responses API and Conversations API, with documented…
AI records and evidence
AI products, models, and the changes that matter.
Find what an AI product does, what it costs, what changed, and the evidence behind the record.
Featured decision tool
Can I Use?
Check an exact AI model, access route, purpose, and region against availability, licence, contract, regulatory, and price conditions—with the primary sources attached.
Model + route
Purpose + region
Licence + contract
Regulation + price
Yes · Conditional · No · Unknown
Check a model and route
Track
Latest intelligence
The launches, changes, and deadlines that matter now.
Launched
RayNeo iO
RayNeo announced the iO smart glasses on August 21, 2026. The official product page labels the device as coming September…
Changed / Deadline
OpenAI Agent Builder shutdown deadline: November 30, 2026
Export and inventory Agent Builder workflows, choose a target architecture, rebuild and regression-test tools, state, guardrails and ChatKit behavior, then cut over with rollback before November 30.
OpenAI GPT Image API shutdown deadline: December 1, 2026
Replace retiring GPT Image model IDs with gpt-image-2 in a test environment and re-baseline generation, editing, transparency, safety, latency and cost before December 1.
Qwen3.8 Max field guide
Qwen3.8 Max: identity, access and deployment limits
A source-led guide that separates the hosted Max product from the open-weight 2.4T-A95B checkpoint.
Read the field guideThree-model field test
Qwen3.8 Max vs Kimi K3 vs DeepSeek V4 Flash
A 45-call comparison that keeps pricing, context, licences and independent-evidence limits visible.
Inspect the comparison24 GB local-agent test
What Muse Glimmer did on a 24 GB Mac
A bounded seven-task comparison that keeps the Mac/Metal and A100/CUDA results separate.
Inspect the test and limitsMuse Glimmer field guide
Muse Glimmer: weights, hardware and limits
A source-audited guide to Meta’s 30B multimodal agent model, local hardware and the limits of launch claims.
Read the field guideGLM-5.3-Flash review
Five real GLM-5.3-Flash tests
Five bounded browser tests, executable-code checks, pricing and explicit limits on what the results establish.
See the tests and limitsGLM-5.3 field guide
GLM-5.3 specs, weights and hardware
A source-audited field guide separating Z.ai documentation, released weights, vendor benchmarks and local deployment math.
Read the GLM-5.3 guideGLM-5.3-Flash identity
What Ox Alpha actually was
The evidence-led identity record for Z.ai’s anonymous Ox Alpha preview and the named GLM-5.3-Flash release.
Inspect the identity evidenceGGUF selection guide
Choose a Qwen3.8-27B quantization without guessing
A source-audited file-selection guide that separates publisher measurements, manifest facts and calculated memory lower bounds.
Choose a quantizationQwen3.8 model field guide
Inspect Qwen3.8-27B specs, evidence and hardware
A source-audited guide to architecture, benchmarks, memory lower bounds, long context and locally observed 24 GB results.
Read the field guide24 GB model comparison
Choose the right 27B-class model for a 24 GB GPU
A reproducible Qwen3.8, Qwen3.6 and Gemma 4 comparison with raw prompts, telemetry, failure handling and source snapshots.
Inspect the comparisonLocal deployment field guide
Choose a MiniMax H3 local runtime
A source-audited comparison of ComfyUI, Diffusers, SGLang, vLLM, hardware tiers, and the hosted boundary.
Read the deployment guideLocal AI setup guide
Run MiniMax H3 in ComfyUI
A file-by-file setup guide that separates official workflow requirements from untested performance claims.
Read the ComfyUI guideAI video control guide
What local MiniMax H3 actually changes
A source-bounded guide to hosted moderation, local control, licensing, consent, and the limits of the word uncensored.
Read the control guideEvidence-led model comparison
Qwen3.8-27B vs DeepSeek V4 Flash
A comparison that keeps official facts, independent tests, local observations, and unresolved evidence gaps separate.
Read the comparisonLocal AI field guide
What it takes to run Qwen3.8-27B locally
A source-audited hardware guide with tested observations clearly separated from estimates and community results.
Read the hardware guideSponsored review
Base44: pricing, security, ownership, and production limits
A clearly disclosed, evidence-bounded review of Base44 for teams deciding whether an AI-built app can move beyond a prototype.
Read the Base44 reviewCoding agent review
Cursor Composer 2.5: what changed, what it costs, and where it breaks
A source-bounded review of Composer 2.5, its pricing model, and a repair workflow for a real failure case.
Read the Cursor reviewOperating guide
Use Codex across app, CLI, IDE, and cloud tasks
A practical map for the ChatGPT desktop app, CLI, IDE, cloud tasks, worktrees, skills, MCP, Scheduled tasks, approvals, sandboxes, CI, and review.
Read the guideResearch note
Did Claude solve the Riemann hypothesis?
No. The reported 67.25% result is an asymptotic lower bound, while the full conjecture remains open.
Read the evidence
Compare
Compare
Compare AI models, products, and operating costs.
Evidence standards
How Kingy verifies
The evidence boundary stays visible from source check to published result.
- Source checkedPrimary documentation is linked.
- Kingy testedObserved workflows are identified.
- Limits statedWhat the evidence cannot prove is explicit.
- Commercial ties disclosedPaid relationships and supplied access stay visible.
Air Jelly
- Product
- A proactive context layer for macOS.
- What Kingy tested
- The proactive-agent workflow, memory behavior, task extraction, reports, and alerts.
- Measured response
- 426,000 public YouTube views in the retained August 16, 2026 snapshot.
- Limitation
- Observed platform response is not proof of attributable sales or ROI; the evidence does not establish long-term accuracy or independently audited privacy behavior.
Disclosure: The Air Jelly video was a paid sponsorship with supplied access.
For Companies
For Companies
Reach AI buyers with evidence-led content and distribution.
Commercial relationships are disclosed, and testing limits stay visible.
For Companies