Kingy Video Test
This $1/Hour AI Model Might Replace Opus
MiniMax M2.5 review companion with official sources, current plan pricing, testing limits and the paid-sponsorship disclosure.
Disclosure: this video was sponsored.
Sources, disclosures, dated pricing observations and chart applicability are documented in the reviewed companion record below.
Video
Full evidence
Kingy Evidence Page · checked 12 August 2026
COMPARE
Use the M2.5 video as dated coding-workflow evidence, not as current price or model-ranking proof. Compare MiniMax’s current M3/M2.7 surface on your exact plan, region, IDE and repository before committing.
Best for: buyers deciding whether MiniMax belongs in a current coding-agent evaluation. Skip this evidence as a purchase shortcut if you need a verified benchmark win, fixed total cost, production concurrency, licensing, or a repeatable repository result.
What the recording actually shows
These are exact automatic-caption boundaries from the 27 February 2026 video. They are historical workflow observations—not a new test run, current-model result or production acceptance.
| Time | Observed workflow | Limit |
|---|---|---|
| 00:39–02:41 | Configured the M2.5 API workflow in VS Code with Kilo Code and the MiniMax provider. | Historical creator demonstration; no instrumented acceptance run. |
| 02:41–04:08 | Generated and reviewed a luxury product-launch website. | Historical creator demonstration; no instrumented acceptance run. |
| 04:08–05:36 | Generated and reviewed a habit and focus tracker app. | Historical creator demonstration; no instrumented acceptance run. |
| 05:36–06:23 | Generated and reviewed a Three.js Empire State Building visualization. | Historical creator demonstration; no instrumented acceptance run. |
Use it, wait, or skip?
Use this receipt
To shortlist the exact kinds of coding outputs Kingy demonstrated and to design a current retest.
Wait
Until your current region, plan, checkout total, model ID, IDE integration and data settings are confirmed.
Skip the headline
If “$1/hour” or “better than Opus” is the deciding claim. Neither is independently reproduced here.
Current state versus the recorded surface
| Decision field | Status | What is defensible |
|---|---|---|
| Model identity | Changed | The video tested M2.5. Fresh MiniMax documentation identifies M3 as latest and lists M2.7 and M2. |
| Price | Unresolved | The headline’s “$1/hour” and the prior $20/$50/$120 observation are not verified as a current buyer total. Fresh HTML is retained; checkout values are client-rendered. |
| Access | Partly supported | Current MiniMax docs describe API and coding-plan access. Exact region, account, quota, taxes and checkout remain buyer-specific. |
| Compatibility | Protocol supported | Anthropic-compatible and OpenAI-compatible formats are documented. Behavioral equivalence, Kilo Code reliability and tool use are not established. |
| Licensing and data use | Unresolved | No commercial-use, training, retention or repository-confidentiality conclusion is drawn from the retained product and pricing pages. |
| Production reliability | Not tested | No concurrency, rate-limit, fallback, function-calling, multilingual, security or repeated-run fixture was retained. |
What failed—or is simply missing
- No raw prompt bundle, request IDs, repository before/after, acceptance suite, token log, latency trace, invoice or failure denominator.
- No matched Claude Opus run under the same prompt, repository, tools, context and output budget.
- No function-calling, multilingual, heavy-concurrency, deployment, persistence, auth, accessibility or security acceptance result.
- No current exact plan total that survives region, quota, taxes and checkout.
Claim ledger
| Claim | Status | Qualification | Sources |
|---|---|---|---|
| The recorded workflow configured MiniMax M2.5 through an API-backed VS Code/Kilo Code setup. | historical_surface_specific_claim_not_current | Recorded 2026-02-27; setup evidence, not a current integration or production-reliability guarantee. | source-005, source-006 |
| The video generated and reviewed a luxury product-launch website. | historical_surface_specific_claim_not_current | A creator demonstration, not a retained repository diff, acceptance suite or matched comparator. | source-005, source-006 |
| The video generated and reviewed a habit/focus tracker app. | historical_surface_specific_claim_not_current | A recorded output inspection; no install, persistence, auth, accessibility or deployment fixture was retained. | source-005, source-006 |
| The video generated a Three.js Empire State Building visualization. | historical_surface_specific_claim_not_current | A recorded visual result; no source bundle, cross-browser run, performance trace or geometry audit was retained. | source-005, source-006 |
| MiniMax’s current model-invocation documentation identifies MiniMax-M3 as the latest M-series model and also lists M2.7 and M2. | supported | Current platform identity only; it does not transfer the M2.5 video result or establish quality versus Claude Opus. | source-001, source-003 |
| Current MiniMax documentation offers Anthropic-compatible and OpenAI-compatible request formats. | supported | Protocol compatibility, not behavioral equivalence, tool reliability or drop-in application compatibility. | source-001 |
| The video headline’s “$1/hour” shorthand and the earlier $20/$50/$120 plan observation are not verified as current buyer pricing. | unresolved | Fresh official HTML is retained but exact checkout values are client-rendered; verify region, taxes, quota and checkout on purchase day. | source-002, source-004, source-007 |
| The video was a paid MiniMax sponsorship; the retained production record says MiniMax provided no-cost product/account/credits/access and no brand editorial approval or control. | supported | Curtis-supplied commercial facts for this exact video. A referral-coded discount URL is present; Curtis reports no affiliate/referral commission. | source-007 |
| The video’s 80.2 SWE-Bench Verified comparison, effective cost, and production performance are not independently reproduced by this receipt. | unresolved | No retained benchmark harness, raw task outputs, denominator, comparator run, invoices or usage trace. | source-005 |
| Function calling, multilingual behavior, heavy concurrency, licensing, data use, rate limits and production reliability remain untested for the buyer’s current configuration. | unresolved | These are explicit test gaps raised by buyer questions or required by the current category method. | None — test required |
The retest that would change the verdict
- Freeze the exact model/endpoint, region, account and plan; IDE/extension versions; repository commit; prompts, tools and data settings.
- Run the same representative repository tasks at least three times, retaining all outputs, failures, request IDs, token use, latency and buyer cost.
- Score machine-checkable requirements, tool-call correctness, schema validity, tests, regressions, recovery and silent routing/fallback.
- Compare only against another model under the same repository, prompt hierarchy, tools, output budget and repetition count.
Buyer FAQ from the public comments
How does M2.5 compare in cost with frontier models now?
Do not use the headline as a current tariff. Check the current MiniMax plan, region, quota and checkout; the fresh retained HTML does not expose a defensible exact buyer total.
Viewer question ID: Ugzy3DCerXmrrLgZvLR4AaABAg. Bounded public sample; not representative polling.
Does it support function calling in production?
This video did not retain a controlled function-calling run. Treat support and reliability as unresolved until an exact current model, tool schema, raw calls and failure log are tested.
Viewer question ID: UgxETDxuzJ-yQc5NkDt4AaABAg. Bounded public sample; not representative polling.
Does it handle multilingual prompts?
Current official copy mentions multilingual programming, but this receipt has no retained multilingual task set or scored output. The buyer-specific result is unresolved.
Viewer question ID: Ugy5NJ6nsMNDR98kEt94AaABAg. Bounded public sample; not representative polling.
Would it perform under heavy concurrency?
No concurrency, throttling, fallback, latency-distribution or rate-limit fixture was retained. Do not infer production capacity from the single recorded workflow.
Viewer question ID: UgwtsuMu05cuULSnxcR4AaABAg. Bounded public sample; not representative polling.
Product, company, article and video joins
- Watch the exact Kingy video
- MiniMax company record
- MiniMax M2.5 article
- MiniMax M2.5 model record: no verified dedicated Kingy model URL; kept as an unlinked label.
- Launch/change record: no verified dedicated public record; current M3 drift is preserved in this receipt.
Official and retained sources
| ID | Class | Source | SHA-256 |
|---|---|---|---|
| source-001 | official | Model Invocation — MiniMax API Docs | 21b8a0bd320a2c31… |
| source-002 | official | Token Plan — MiniMax API Platform | 89128d9fd008906f… |
| source-003 | official | MiniMax CLI — MiniMax API Docs | 53b619587c2262a0… |
| source-004 | official | Coding Plan — MiniMax API Platform | 74724c5887c7b47d… |
| source-005 | affiliate | Retained public video metadata, description and bounded comments | a0c19986f649d9be… |
| source-006 | commercial | Automatic English captions for the sponsored video | d1da8c39b721054f… |
| source-007 | adjacent | Current production B01 content and package state | 36d5f332b9e85ff6… |
Evidence checked 2026-08-12T23:12:00Z. If a source, plan, model or result changes, send a correction; Kingy preserves the dated observation and appends the change.
Have a product or workflow that should face this test?
Production or distribution can be funded. Verdicts, rankings, claim status and corrections cannot.