Gemini 4 Carbon vs Beam 1.0

Gemini 4 Carbon 83

Model reading for gemini-4-carbon on the Gemini API. Platform facts are the Gemini API listing's (gemini-api, re-read 1 November), carried over as RULES.md allows for a model on an already graded platform. Released to developers on 10 November 2026, with the platform's changelog (30). Deprecations with earliest dates and no stated minimum notice (4 of 12). Shutdowns in the last 90 days kept to their table (4 of 8). Dated changelog (15). SDK replies not sampled (5 of 10). Current SDKs (15). Runtimes stated (10). 30 + 4 + 4 + 15 + 5 + 15 + 10 = 83.

Beam 1.0 50

Model reading. The weights were published on 16 October 2026, the date on the Hugging Face model page (the repository was created on 7 October); later commits to 30 October add evaluation results and a guide, which record no model change (30). One model, with a shutdown_date field and a lifecycle section in the docs but no notice period for retirements, and no history of releases or retirements yet (4 of 20). No changelog. Support through the platform and developers@reflection.ai for higher limits, and the model card names reports@reflection.ai. The Hub page shows 15 discussions (10 open, 5 closed), replies not sampled (6 of 25). No first-party SDK, the docs rely on the OpenAI SDKs (5 of 15). Reflection's serving guides for vLLM, SGLang, TokenSpeed, Transformers and Dynamo, a Beam Verifier repository, and fine-tuning recipes now published for NeMo AutoModel, Megatron Bridge, Axolotl and TRL (5 of 10, partial kept). 30 + 4 + 6 + 5 + 5 = 50.

Gemini 4 Carbon 93

Model reading for gemini-4-carbon on the Gemini API. Platform facts are the Gemini API listing's (gemini-api, re-read 1 November), carried over as RULES.md allows for a model on an already graded platform. Function calling, with forced calls by FunctionCallingConfig on the API (20). Structured output (13 of 15). Implicit caching with durable caching on by default, cached input 95% off, explicit caches not supported, per the Cloud guide read on 10 November (15). 1,048,576 tokens shared between input and output (15). Batch at half price on the platform (10). Python and JavaScript SDKs (10). Error codes with backoff guidance (10 of 15). 20 + 13 + 15 + 15 + 10 + 10 + 10 = 93.

Beam 1.0 67

Model reading (tool use, structured output, caching, context, batch, SDKs and defaults, errors). tool_choice of auto, none, required or a named function, parallel_tool_calls, strict tools and up to 128 tools. The tool-calling loop needs reasoning_content sent back on every assistant turn, and the docs say the model sometimes returns a tool-call step without it, after which every request in the session fails (12 of 20). json_object and strict json_schema (15). Cached prompt tokens are reported in usage, but caching can't be steered, the prompt_cache_* options are ignored and no cached price exists in the beta (5 of 10). 1,048,576-token context and 131,072 output tokens (15). No batch endpoint (0 of 10). No Reflection SDK, the OpenAI Python and TypeScript SDKs are documented, and defaults cost an agent turns, with reasoning always on at max, stop and seed ignored and a default output budget not reduced to fit the prompt (5 of 10). Machine-readable error.code values with param, x-should-retry, Retry-After and x-request-id (15). 12 + 15 + 5 + 15 + 0 + 5 + 15 = 67, per the 17 October reading of the developers guides, not re-read on 4 November beyond the models and rate-limits pages.

Gemini 4 Carbon 72

Model reading for gemini-4-carbon on the Gemini API. Platform facts are the Gemini API listing's (gemini-api, re-read 1 November), carried over as RULES.md allows for a model on an already graded platform. Closed service with clear terms, SDKs Apache-2.0 (15). Paid-tier data isn't used, abuse retention has no number, and request storage is on by default for new paid projects (20 of 30). Deprecations table (15 of 20). Data may be cached anywhere Google has facilities; no subprocessor list checked (5 of 20). Every benchmark figure is Google's own. 15 + 20 + 15 + 5 = 55.

Beam 1.0 57

Weights under Apache-2.0, an OSI licence, per the model card, which says the licence doesn't restrict distillation or training on outputs. The hosted API is a closed service whose commercial terms are on reflection.ai and can no longer be read (25 of 30, kept). Data handling 10 of 30, partial kept. What remains readable is the API reference's retention header (zero, standard, contributor) and the rate-limits page's request for data retention requirements; the terms, privacy policy, API safety page and zero-retention page are struck. Deprecation 4 of 20 (was 6). A shutdown_date field and a lifecycle section that says to move off a model before that date, with no notice period or dated notice; the fee-increase notice came from the struck terms. Subprocessors 0 of 20 (was 18). The list read on 17 October is on reflection.ai and struck, and no readable page names subprocessors or data locations. 25 + 10 + 4 + 0 = 39.

Gemini 4 Carbon 40

Model reading for gemini-4-carbon on the Gemini API. Platform facts are the Gemini API listing's (gemini-api, re-read 1 November), carried over as RULES.md allows for a model on an already graded platform. Status page at aistudio.google.com/status with no readable components or history (10 of 20) and no readable incident record (5 of 30). Rate-limit tiers and spend limits published, per-model limits in AI Studio (5 of 15). Backoff guidance for 429 and 503 and SDK retries (15). No SLA (0). Carbon's launch stage isn't stated in Google's developer post, the Antigravity post or the Cloud developer guide read on 10 November, and the model ID carries no preview label, so half (5 of 10, a judgement call; Argon's GA earned the platform 10). 10 + 5 + 5 + 15 + 0 + 5 = 40.

Beam 1.0 52

Hosted lines, for the Reflection API. Status page at status.reflection.ai on incident.io, with API, platform and developer docs components and Atom and RSS feeds (20). In the 90 days to 4 November 2026 its incident pages list two full outages of the API component, a serving cluster down from 22:01 to 23:20 UTC on 14 October and elevated errors from 06:10 to 07:10 UTC on 15 October, each an hour or more, so two majors. Three more are labelled partial outages (16 October 02:28 to 02:38 UTC, 22 October rate-limit errors in two windows, 3 November about 20:10 to 21:38 UTC on one inference provider), read as minor. Two majors give 5. Rate limits are described per organisation in requests and tokens per minute and per day, with concurrency caps and headers, but no figures are published (5 of 15). 429 and 503 responses carry Retry-After, daily limits add x-should-retry: false, and the errors page says which codes to retry, with exponential backoff and jitter. There are no idempotency keys, and a chat completion is the billed call, so this run's value for retry guidance without idempotency keys gives 12 (was 15). No SLA found (0). The developers docs read on 4 November carry no beta or preview label for the API; the earlier beta reading came from reflection.ai pages now struck (10). 20 + 5 + 5 + 12 + 0 + 10 = 52.

Gemini 4 Carbon 55

Model reading for gemini-4-carbon on the Gemini API. Platform facts are the Gemini API listing's (gemini-api, re-read 1 November), carried over as RULES.md allows for a model on an already graded platform. Keys bound to a service account and restricted to the Gemini API, with no scopes and a token accepted in the URL (15 of 30, the platform's deduction). Carbon is on the paid tier only, and Google's terms say paid-tier content isn't used to improve products; AI Studio use under a Google AI Ultra subscription falls under terms we didn't read, so 15 of 20 (a judgement call; the platform's 10 is for its free tier). Abuse retention with no number, and new paid projects store requests by default since 26 October (5 of 15). Request logs for billed projects (15). security.txt and a reward programme, no certification naming the API (5 of 20). 15 + 15 + 5 + 15 + 5 = 55.

Beam 1.0 45

Model reading. API keys per project, revocable, shown once and disabled when their creator leaves, with no scopes, expiry or rotation API found (20 of 30). Training 10 of 20, partial kept. The API reference says each response's reflection-data-retention header reports contributor when content may be used to train Reflection's models "under contributor terms your organization accepted". The API safety page and commercial terms read on 17 October are struck (reflection.ai now disallows our crawler), so whether training is off by default can't be read. Retention 5 of 15, partial kept. The header can report zero, standard or contributor, and no readable page states periods or whether zero retention is offered. x-request-id on every response and the retention header, with usage dashboards behind the login and not read (5 of 15). Security programme 5 of 20 (was 6). The model card names reports@reflection.ai for security vulnerabilities; developers.reflection.ai has no security.txt (404 on 17 October), and the safety report and reflection.ai's security.txt can't be read. No bug bounty, SOC 2 or ISO 27001 found on a readable page. 20 + 10 + 5 + 5 + 5 = 45.

Gemini 4 Carbon 20

Model reading for gemini-4-carbon on the Gemini API. Platform facts are the Gemini API listing's (gemini-api, re-read 1 November), carried over as RULES.md allows for a model on an already graded platform. No machine payment protocol (0). Introductory prices published in Google's 28 October announcement without a login, $2 and $10, then $4 and $20; the Gemini API pricing page as read on 10 November showed no Carbon row, so the figures rest on the announcement (20). No free tier: Google's developer post puts Carbon on the paid tier (0). A Google account and AI Studio for a key (0). 0 + 20 + 0 + 0 = 20.

Beam 1.0 10

Graded on the Reflection API. No x402, MPP or L402 (0). No per-token, plan or allocation price is published on any readable page; the beta pricing page on reflection.ai is struck (0, was 10). Free tier 10 of 20, partial kept. The rate-limits page describes a daily token allowance and asks those requesting higher limits whether they are "interested in paid usage", and the authentication page says some organisations must verify a credit card before keys work. A person signs up in the browser at platform.reflection.ai to get a key (0). 0 + 0 + 10 + 0 = 10.