August 14, 2026
Gemini 3.1 Pro is live at 16.20 per million tokens — the first Gemini Pro tier on Kyma — and Nano Banana 3 Flash is repaired after its upstream preview endpoint stopped answering
Kyma now serves a Gemini Pro tier. Until today every Gemini in the catalogue was a Flash or a Flash Lite, so anything those tiers could not carry had to leave the family entirely.gemini-3.1-pro fills that gap: a 1M-token context, a 64K output ceiling, and text, image, audio, video and file input on the same request.
It lists at 16.20 out per million tokens. Above 200,000 prompt tokens the rate steps to 24.30, because that is what it costs us there — the threshold is on the prompt, so you know which rate applies before you send. Google ships this as a preview build, and the model page says so; test before you put it on a critical path.
Separately, Nano Banana 3 Flash is fixed. It was routed to Google’s preview image endpoint, and that endpoint stopped resolving — every request to it was failing. It now runs on the generally available endpoint. The model id you call is unchanged and so is the price.
- Reach for a Pro tier without leaving Gemini — hard reasoning and long-horizon agentic work at 1M context, with the same OpenAI-compatible call and the same automatic failover as everything else
- Know the long-prompt rate up front — the 200K step is published next to the base rate on
/pricingand in/v1/models, not discovered on the invoice - Nano Banana 3 Flash answers again — the upstream endpoint swap was transparent: same
nano-banana-3-flashmodel id, same $0.061 per image
Gemini 3.7 Flash is live at 5.06 per million tokens — and half that, 2.53, while the launch promotion runs
Google shipped Gemini 3.7 Flash on 13 August and it is callable on Kyma now. It takes a 1M-token context like the rest of the Flash family, and raises the output ceiling from 8,192 tokens to 65,536 — the reply, not the prompt, is what got eight times bigger. It is on sale. The list price is 5.06 out per million tokens; while the launch discount holds upstream we publish 2.53, and the end date is on the model in/v1/models. The list price is the number the sale returns to, so nothing rises unannounced when it ends.
- Stop chunking long replies — a full document translation or a 40,000-token structured extraction now fits in one response instead of a loop that stitches pieces together
- Send video, audio and files to one model — the widest input set in the Flash family, with tool calling, reasoning and structured outputs on the same request
- Check what you actually pay before you commit —
/v1/modelspublishes the promotional rate, the list rate it returns to, and the date it ends, so a budget built on it does not go stale silently
7 prices fall and 14 rise — every one computed from what the request actually costs to serve
7 models get cheaper.
14 models go up, because they now cost more to serve than their prices covered.
Prices are computed from real serving costs — they can fall, and never rise without notice.
Learn more: Pricing
/v1/models now quotes the price you actually pay today, so a model on promotion no longer reads as twice its real cost
When promotional pricing shipped,/v1/models kept returning the list price under pricing and put the sale in a separate promotion field. For the four GPT-5.6 models on offer, that meant the endpoint advertised twice what it charged unless you knew to read a field released the same day.
pricing is now the rate in force — the promotional one while a promotion is running, the list rate otherwise. A new list_price field carries the stable rate a promotion returns to, and it is always present, equal to pricing when nothing is on offer, so you can read one field unconditionally and never branch. promotion is unchanged.
Billing never changed and was never affected: a bill follows the upstream invoice, which already carries the discount.
Learn more: Models List API
August 3, 2026
You are now billed at the exact rates published by the suppliers, eliminating incorrect pricing on several popular models.
We corrected twenty-seven cost rows that were previously estimated by hand and lacked a verified source. Several models had incorrect pricing that did not match the official supplier rates, which has now been aligned with the published catalogs. Learn more: Pricing Catalog API · PricingThe model uptime dashboard now accurately reflects supplier availability instead of showing false outages caused by our own probe limits.
Our health probes were occasionally hitting our own account limits, resulting in payment required errors that were incorrectly counted as model downtime. We updated the health rollup to classify these specific authentication and billing errors as unmeasured, ensuring the public uptime metrics only reflect actual supplier availability. Learn more: overview on KymaAugust 2, 2026
1 model got cheaper to call, effective immediately
One model gets cheaper.
Prices are computed from real serving costs — they can fall, and never rise without notice.
Learn more: Pricing
August 1, 2026
You can list everything one lab makes in a single request
GET /v1/models?creator=anthropic returns that lab’s models and nothing else.
Pass several — ?creator=anthropic,google — to compare labs side by side.
?owned_by= is accepted as the same filter, so you can filter by exactly the
value you just read in the response.
Asking for a lab Kyma does not carry returns an empty list rather than the whole
catalogue, so a typo tells you it was a typo.
- Compare two labs —
?creator=anthropic,googleputs them side by side before you commit to one - Audit a family — see every model one vendor makes, and what each costs, in a single call
- Build a vendor picker — group your model dropdown by lab without hard-coding the list
One click unsubscribes you from Kyma email, and it sticks
Every Kyma email carries a one-click unsubscribe that your mail client can act on directly, and the suppression is honoured across all future sends rather than per-campaign.You can add live web results to 25 models by appending one suffix to the model id
:online works on 25 models instead of 2. Call deepseek-v4-pro:online and the
model answers with current web results; drop the suffix and it does not.
No separate endpoint, no second request, no search vendor to sign up with.
- Answer from today — ask about a release that shipped this morning and get the release, not the training cutoff
- Keep your model — add
:onlineto the model you already tuned prompts for instead of switching to a search-native one - Turn it off per request — drop the suffix when the answer should come from the model alone, with no extra call
Grok, Meta’s newest line and Claude are all callable with the same key as everything else
The Grok family joins the catalogue, alongside Meta’s Muse Spark 1.1 and Llama 4 Maverick, and eight more models filling out families Kyma already sells. Claude is listed publicly. Listing Grok exposed four defects in how tool support was advertised, and those were fixed before it shipped rather than after.- Try a frontier model without a new account — Grok and Claude bill against the key you already have
- Swap in one line — change the model id; the request shape, the fallbacks and the receipts stay the same
- Compare on your own traffic — run the same prompt across labs and read the cost back from
usage
Published prices follow the cheapest route that can serve you, not whichever one happened to carry the last request
Nineteen prices were recomputed from real traffic on this day. The published price is now anchored on the cheapest route that can serve a model, so a cheaper route arriving lowers what you pay rather than changing nothing you can see. Learn more: PricingWhat a model says it can do now matches what its provider says it can do
Six models advertised capability flags that disagreed with the provider’s own specification — tool use, structured outputs and vision among them. A flag that is wrong in the optimistic direction fails at request time, on your side. Every flag in the catalogue is now checked against the provider spec on a schedule, and a disagreement fails the build. Learn more: Models List APISearching the models page no longer throws you somewhere else on the page
Filtering 99 models down to a handful made the page much shorter, and the browser had to put your scroll position somewhere — usually a long way from what you were reading. It looked like the page had reloaded. Results now stay where you can see them.July 31, 2026
You can compare models on measured speed and uptime instead of taking a claim on trust
Every model in the catalogue now shows Throughput, Latency and Uptime — measured, not quoted. The same prompt goes to every model through this API every six hours, so the numbers compare with each other rather than with a vendor’s benchmark. Uptime is every observation over 30 days, probes and real customer requests together, and counts a request as successful only if the model you asked for is the model that answered. A dash means we have not measured enough yet, and says so instead of guessing. The table also filters by capability, use case and creator, and sorts on any column.
- Pick on evidence — sort by throughput or uptime instead of trusting a launch post
- Rule a model out — a dash means it has not been measured enough, and says so rather than guessing
- Check before you migrate — see how a model has actually behaved over 30 days, not on the day it launched
When a supplier cuts its price, yours falls too — automatically, and within a day
Published prices are recomputed from what serving actually costs, in both directions. A cost that falls pulls the price down; a cost that rises can only push it up with notice. Long prompts are priced in steps, matching how suppliers charge above a threshold, so a large request is billed on the rate that actually applied to it rather than a flat average. Twenty prices were recomputed on this day. Learn more: PricingYour account is harder to attack, and a database read no longer yields anything replayable
Two changes to how sessions and sign-in work. Repeated failed logins are now rate-limited per account and per IP on the path production actually serves. Only failures count — signing in normally is never throttled, which matters where many people share one address. Session tokens are stored as a digest rather than in the clear, so what is at rest cannot be pasted into a request. Existing sessions were not invalidated; nobody was signed out. Learn more: Authentication guideThe cost in your streaming response is what you were charged, not what your request cost Kyma
usage.cost on a streamed response was not the amount billed to you. The number
was real, it was simply the wrong side of the transaction, and it could differ in
either direction.
Non-streaming responses were already correct. Both now report your price.
Learn more: Chat Completions API · Pricing
July 30, 2026
Older clients that speak the legacy completions shape work without changing your code
POST /v1/completions accepts the pre-chat request shape and translates it. It
is an adapter over the same routing, billing and never-die behaviour, not a
second pipeline — so a legacy client gets the same fallbacks and the same
receipts as everything else.
- Keep an old client — point it at Kyma without rewriting it for the chat shape
- Migrate on your own schedule — move to
/v1/chat/completionswhen it suits you, not because the API forced it - Get the same guarantees — an adapter over the same routing, so a legacy call still gets never-die fallbacks
You can build retrieval on Kyma without a second vendor for the embedding half
POST /v1/embeddings is live, OpenAI-compatible, billed on input tokens only —
there is no output side to charge for.
RAG needed two things and Kyma only had one of them. Now the retrieval half and
the generation half come from the same key.
- Index a corpus — embed your documents with the same key that answers questions about them
- Drop a vendor — retrieval and generation stop needing two accounts, two bills and two outage pages
- Pay for input only — there is no output side on an embedding, and you are not charged for one
Every model has its own page, with the whole price and a plain-markdown mirror an agent can read
Each model now has a page carrying its full price — input, output and cached, not a rounded headline — plus context window, capabilities and what it is good for. Every page has a.md twin at the same path, so an agent fetching it gets the
facts without parsing a web page.
- Read the whole price — input, output and cached, not a rounded headline
- Feed it to an agent — every page has a
.mdtwin at the same path, so a crawler gets facts, not layout - Check before you call — context window, capabilities and what it is good for, in one place
You are billed for the route that actually served you, and never above the published price
A model can be served by several suppliers at different costs. Billing used to assume one of them. It now records the endpoint that served, prices from that, and caps the result at the published price — so a fallback to a dearer route is Kyma’s cost to absorb, not yours. Learn more: PricingPages load faster, and the site stops rendering dark on a light theme
113KB of analytics that nothing used came off every page, and session replay moved off the critical path. Separately: choosing “system” was being read as “dark”, so the header rendered dark on every light page. It now follows the system.July 29, 2026
You have until 20 October to move off nano-banana, and Kyma stopped recommending it today
The endpoint behindnano-banana shuts down on 2026-10-20. It still serves
until then and nothing changes for existing calls.
It is no longer recommended anywhere in the catalogue or the docs, so nobody
starts a new project on a model with a date on it.
Learn more: nano-banana on Kyma
Six models joined the catalogue, including the cheapest one Kyma sells
- Kimi K3 — Moonshot’s 2.8T open-weight reasoning model, 1M context, vision in
- Gemini 3.6 Flash and Gemini 3.5 Flash Lite
- Qwen 3.7 Flash — the cheapest model in the catalogue
- GPT-5.6 Terra, priced on list rather than on a promotion that expires
- Gemini 3.1 Flash Live, the successor to the 2.5 voice SKU
- Reach for 1M context — Kimi K3 and Gemini 3.6 Flash both take a repository-sized conversation
- Cut the floor price — Qwen 3.7 Flash is the cheapest model in the catalogue
- Move a voice workload — Gemini 3.1 Flash Live succeeds the 2.5 voice SKU that is retiring
Five models were repriced against what they actually cost to serve — two of them had been sold below that
qwen-3.7-max and gemma-4-31b were priced under what the route serving them
charges. Three more were re-anchored against the dearest route in their chain
rather than whichever supplier happened to be written first.
Prices are computed from real serving costs. They can fall, and they do not rise
without notice.
Learn more: Pricing · qwen-3.7-max on Kyma
Multi-turn reasoning with Gemini 3 keeps its train of thought across turns
Gemini 3 returns a signature with its reasoning that must come back on the next turn. Kyma was dropping it, and the failure was invisible: the model simply reasoned from scratch each turn, and the answer looked plausible. The signature now survives the round trip on both paths, and a request that cannot carry it fails loudly instead of quietly degrading. Learn more: Tool Calling guideJuly 28, 2026
A long conversation is never rescued by a model too small to hold it
Never-die routing tries other models when one fails. It did not check whether the replacement could fit the conversation, so a 200k-token request could be handed to a model with a 32k window and fail a second time — slower, and for a different reason than the first. Fallbacks are now filtered by context window before they are attempted. Learn more: Routing guideJuly 27, 2026
You can ask the catalogue for exactly the models you need instead of reading all of them
GET /v1/models now filters on what a model is and what it is for:
?type=image,video, ?tags=tool-use,vision, ?use_cases=write-code.
Several values in type mean “either” — a model has one type. Several in tags
or use_cases mean “all of them”, because a model has many. Both serving paths
answer identically.
Learn more: Models List API
An error now tells you whose problem it is, so you know whether retrying will help
When an upstream failed, its status code came back as though Kyma had produced it. A 429 from somewhere else looked like your rate limit; a 401 looked like your key. Errors Kyma owns now carry Kyma’s status. And when a request is served through an alias, the response says which one resolved, soX-Kyma-Model and the id you
asked for can be reconciled.
Learn more: Error Handling guide
June 2026
Every model got a page with live numbers instead of a specification copied from a launch post
Model detail pages now carry measured data — what the model costs, what it supports, and how it has actually behaved — rather than a static description written on the day it was added. Rankings rows became clickable through to them, so a model that catches your eye in a table is one click from its own page.- Check before you switch — read a model’s real numbers rather than its announcement
- Follow a name from the rankings — the table rows lead somewhere now
- Compare two pages — the same fields, in the same order, for every model in the catalogue
Six models joined, including two flagship coding models and the strongest open-weight tier Kyma had carried
GLM-5.2 and Kimi K2.7-Code arrived as flagship coding models. Alongside them: qwen-3.7-plus, minimax-m3, nemotron-3-ultra-550b and step-3.7-flash, plus ElevenLabs v3 for speech and a Gemini 3.5 Live Translate voice SKU.- Put a coding model behind an agent — GLM-5.2 and Kimi K2.7-Code were added for exactly that shape of work
- Reach 550B open weights — nemotron-3-ultra is the largest open-weight model in the catalogue
- Translate in a live call — the Live Translate SKU runs on the realtime path, not as a batch job
Speech and transcription stopped failing outright when one supplier did
Chat has had four layers of never-die routing for months. Audio had none: one supplier failing meant your request failed. Transcription now retries, then falls back to a second model, then a third. Speech gained cross-pool failover on a rate limit rather than surfacing it to you. Both paths behave the same on either backend, which had to be checked rather than assumed. Learn more: Audio Transcriptions API · Routing guideJune 19, 2026
2 new flagship models: GLM 5.2 and Kimi K2.7 Code
Two new open-weight coding models, both live-verified across their serving paths before launch:glm-5.2— Zhipu’s newest frontier-open flagship. 744B MoE, 1M-token context (up from ~200K on GLM 5.1), #1 open-weight on the Intelligence Index v4.1. 5.94 per 1M tokens. Theglm-flagshipalias now resolves here.kimi-k2.7-code— Moonshot’s coding specialist. +21.8% on Kimi Code Bench v2 vs K2.6 with ~30% fewer reasoning tokens, 262K context, text+image input, always-on thinking. 5.40 per 1M tokens.
glm-5.1 and kimi-k2.6 are still callable by model ID, and agent/best-agent still resolve to kimi-k2.6.
June 10, 2026
4 new models: Qwen 3.7 Plus, MiniMax M3, Nemotron 3 Ultra, Step 3.7 Flash
Four new text models, all with tool calling and live-verified before launch:qwen-3.7-plus— successor to qwen-3.6-plus. 1M context, adds vision, 2.16 per 1M tokensminimax-m3— agentic coding focus, 1M context, text+image+video input, 1.62 (same price as M2.5)nemotron-3-ultra-550b— largest US open-weight model (550B MoE), 1M context, 3.375step-3.7-flash— cheap multimodal flash tier, 256K context, tool calling, 1.553
- Reliability: per-attempt upstream timeout — a hung provider now fails over instead of hanging your request
- Billing hardening: settle idempotency, usage estimation fallback, and time-to-first-byte now measured on every request
deepseek-r1input price adjusted to $0.74/1M (output unchanged)
June 4, 2026
ElevenLabs v3 — most expressive TTS + low-latency streaming
AddedModeleleven-v3— the most expressive text-to-speech model on Kyma. Audio tags and emotional range for character voices and dialogue, lifelike delivery across 70+ languages. Same$0.405/1K charas Multilingual v2. Call it onPOST /v1/audio/speechwith any ElevenLabs voice.
FeatureOpt-in low-latency TTS streaming — pass"stream": trueonPOST /v1/audio/speechto receive audio progressively as it’s synthesized. Time-to-first-audio drops to ~0.4s (from ~1.8s). The response is still a single completeaudio/mpegstream, so existing clients keep working unchanged. Currently applies to the MiniMax speech models.
May 2026
Speech, music and sound effects arrived, so an audio app no longer needs a second vendor and a second bill
May was the month Kyma stopped being text-and-pictures. Text-to-speech landed first, then music, then sound effects, then voice cloning and voice design — each as an ordinary endpoint behind the key you already had.GET /v1/audio/voices lists every voice across the catalogue in one call, so a
picker does not have to know which house a voice comes from.
- Narrate something long — send a chapter to
/v1/audio/speechand get a file back, billed per character - Score a video —
/v1/audio/musicwrites to a prompt and bills per second, so a 30-second bed costs what 30 seconds costs - Clone a voice once and reuse it — upload a sample, get an id, and pass that id anywhere a voice is accepted
Image and video generation grew a real catalogue, and pricing that matches how each one actually charges
Recraft v4 arrived with a print-ready 4MP tier and native SVG output, alongside flux-2-pro, gpt-image-2 and MiniMax’s image-01. Video came with Hailuo 02 at three resolutions. Pricing grew to match: some of these bill per megapixel, some per image, some per second of output. The catalogue now prices each one the way its supplier does, instead of flattening everything into a token count that never applied.- Ship vector art — Recraft v4 Vector returns SVG, not a raster you have to trace
- Print at 4MP — the Pro tier exists for output that has to survive being printed
- Generate video by the second — Hailuo 02 at 512p, 768p or 1080p, billed for what it produced
Web-grounded answers became callable models rather than a separate search product to integrate
Perplexity Sonar and Sonar Pro joined the catalogue. They answer from the live web and cite what they used, through the same request shape as every other chat model. They also introduced a billing primitive Kyma did not have: a per-request fee on top of tokens, because that is how search is actually charged. It is shown separately rather than folded into a token rate that would misprice short and long requests in opposite directions.- Ask about this week — a question whose answer changed yesterday gets yesterday’s answer
- Keep one client — same endpoint, same key, same fallback behaviour as the rest of the catalogue
- See the search fee — the per-request charge appears as itself, not hidden in a token price
Signing up got harder to abuse and easier for real users on shared connections
Signup gained a Turnstile challenge and email verification before free credit is granted, which closed four ways the rate limit could be walked around. The same month fixed the opposite problem. A large share of Vietnamese traffic sits behind carrier-grade NAT, so many genuine users present one address: the per-IP signup cap went from 10 to 30 per day, Google sign-in stopped counting against it, and the login limiter was changed to count only FAILED attempts. Signing in correctly has never been the thing worth throttling. Learn more: Authentication guideMay 17, 2026 (later)
Google media models — 7 new SKUs + public pricing catalog
Google’s media models (Imagen 4, Veo 3, Nano Banana) are now available on Kyma. All seven SKUs ship behind the existing/v1/images/generations and /v1/videos/generations endpoints — no client change needed beyond picking the new model value.
Added — image models
Modelimagen-4-fast— Google Imagen 4 fast tier, $0.027/imageModelimagen-4— Google Imagen 4 default, $0.054/image (recommended)Modelimagen-4-ultra— Google Imagen 4 print-ready, $0.081/imageModelnano-banana— Gemini 2.5 Flash Image with native edit-mode (image-in + prompt → image-out), $0.046/imageModelnano-banana-3-flash— Gemini 3.1 image preview, $0.046/image
Modelveo-3-fast— Google Veo 3 fast tier, 720p no-audio, $0.135/sec (recommended budget)Modelveo-3— Google Veo 3 flagship, 1080p with native audio (dialogue + ambient + lip-sync), $0.540/sec
APIGET /v1/pricing— full catalog: text + image + video + audio in one round-trip. Replaces the partial/v1/credits/pricing(kept for backward compat). Cache-Control 60s.APIGET /v1/limits/tiers— full 5-tier matrix, public, cache-Control 300s. Drives the Rate Limits guide and dashboard signup flow.
May 17, 2026
Audio infrastructure refresh
Eight changes shipped over two days. Headline: realtime audio now serves up to 5000 concurrent sessions per project (a 100× lift), and STT gains a URL-fetch mode + automatic fallback. AddedFeatureRealtime audio WebSocket proxy — scales to 5000 concurrent sessions (was 50). See Realtime Audio.FeatureSTT URL mode —POST /v1/audio/transcriptionsaccepts JSON{"audio_url": "https://..."}up to 100 MB. Multipart upload remains capped at 25 MB. See Audio Transcriptions.FeatureSTT never-die failover — when the primarywhisper-v3-turbohits a transient hiccup, Kyma transparently retries, then routes to a timestamp-preserving secondary (whisper-1), and for plain-text transcripts a final tertiary (gemini-3-flash-audio) — same request, same price. Response carriesX-Kyma-Fallback(serving model) andX-Kyma-Fallback-Layerwhen it fires. See Audio Transcriptions.FeatureTier override for heavy users — partners and enterprises can request Tier 4 limits without the $1000 lifetime deposit. See Rate Limits — Need higher limits?.
UpdateAudio rate limits split into per-capability sub-pools (transcription, understanding, speech). Saturating one no longer blocks the others. See Rate Limits — Audio limits.UpdateAudio concurrency caps raised across all tiers — Tier 4 now 100 total audio slots (was lower).Updateminimax-music-probacked bymusic-2.6(wasmusic-2.5). API contract and pricing unchanged.
OpsIn-process audio load_factor monitor with Telegram alerts when sustained utilization exceeds 0.70 / 0.85 thresholds.
May 1, 2026
- [Model]
gpt-image-2— OpenAI’s flagship image model live on Kyma. Near-perfect text-in-image (multilingual: Japanese, Korean, Hindi, Bengali), reasoning-augmented composition, photoreal output. Quality dropdownlow | medium | high: $0.014 / $0.081 / $0.297 per 1024² image. 4 sizes supported (1024², 1024×1536, 1536×1024, 2048²). Single SKU, no-proderivative — keeps OpenAI’s exact model ID. - [API] Unified picker taxonomy across image, video, and audio composers. Three top-level tiers (
Quality / Fast / Cheap) replace the prior ad-hoc grouping; capability sub-axis surfaces SVG (image) and Speech / Music / SFX / STT / Audio-understand (audio). Hailuo 02 1080p moved Quality → Fast; Quality is now reserved for SOTA-class output (Kling 3 Pro, Seedance 2 Pro, gpt-image-2, flux-2-pro, ideogram-v3, recraft-v4-pro)./v1/modelsexposes newtierandcapabilityfields with?tier=qualityand?capability=vectorfilter params. Backward-compat:cost_tier,quality_tier,latency_tierstill returned. - [Pricing] New
per-qualitypricing mode inIMAGE_COSTS. Hold and settle thread thequalityparam end-to-end so aquality=highrequest books the right amount on the hold (no refund-and-rebill drift on finalize). - [Reliability] Recovery for multimodal jobs after worker death. Hailuo / Image-01 / gpt-image-2 jobs that exceeded their poll budget after the worker died were leaking forever in
processingand holding credits hostage; they now refund cleanly within ~12 minutes for video and ~7 minutes for image. Heartbeat ticker (~1.5 s) keeps in-flight OpenAI calls visible to the sweep so multi-minute high-quality requests don’t get prematurely failed. - [Fix]
not_multimodalvalidator allowlist now reads from a singleMULTIMODAL_PROVIDERSset (was a hardcoded chain). Adding a new model source is one line — same regression that bit MiniMax onboarding before. - [Fix] Audio composer prompt counter NFC-normalizes Vietnamese (and other stacked-diacritic) text before measuring length. Telex-encoded Vietnamese was inflating UTF-16 code units 3× and tripping false
prompt_too_longrejections at ~660 visible characters on 2000-char-cap models. - [Fix] Per-model audio character limits matching upstream provider caps (MiniMax music: prompt ≤ 200 / lyrics ≤ 600; ElevenLabs music prompt ≤ 2000; SFX prompt ≤ 500). Errors now name the SKU and which field overflowed.
April 30, 2026
MiniMax bundle — 9 new SKUs across audio, image, video
Audio, image, and video coverage all expanded under MiniMax’s PAYG pricing — typically 2× to 90× cheaper than comparable hosted equivalents at matching quality tiers. Audio (TTS + music + voice services):- New model
minimax-speech-hd— $0.140/1K char — production multilingual voice, ~2.9× cheaper thaneleven-multilingual-v2. - New model
minimax-speech-turbo— $0.090/1K char — lowest-latency voice on Kyma, ~2.2× cheaper thaneleven-flash-v2-5. - New model
minimax-music— $0.045/song flat — Music-2.0 family, ~90× cheaper thanelevenlabs-musicfor non-hero tracks. - New model
minimax-music-pro— $0.210/song flat — Music-2.5+ richer arrangements at production fidelity. - New endpoint
POST /v1/audio/voice-clone+ model — $2.10/voice flat — clone from 10s-5min reference audio (multipart). - New endpoint
POST /v1/audio/voice-design+ model — $4.20/voice flat — generate a voice from a text description, no reference needed.
- New model
minimax-image-01— $0.005/image flat — cheapest image SKU on Kyma, ~11× cheaper thanrecraft-v4.
- New model
hailuo-02-512p— $0.140/clip — cheapest video tier on Kyma, ~4× cheaper than Kling 2.5 Pro at 6s. - New model
hailuo-02-768p— $0.420/clip — mid tier, balanced quality vs cost. - New model
hailuo-02-1080p— $0.780/clip — full HD hero output, less than half the cost of Kling 3 audio at 10s.
per-song (music), per-call (voice services), per-video (Hailuo) — all flat per request, no duration metering. Image flat-mode SKUs gained an optional listPrice override for safety-buffer rounding.
Voice ID ownership: Cloned and designed voice IDs are gated per Kyma user (migration 064-minimax-voice-clones.sql). Sharing a voice_id with another account returns 403 voice_not_owned from /v1/audio/speech.
Image catalog refresh — 5 new SKUs
The image lineup grew from 4 → 9 active SKUs. Better defaults, cheaper hero shots, native SVG output.- New model
recraft-v4— $0.054 — replacesrecraft-v3as the daily default. #1 on the HuggingFace Text-to-Image Arena, beats Midjourney V8 / DALL-E 3 / FLUX in human preference. Same price as V3. - New model
recraft-v4-pro— $0.338 — V4 quality at 4MP for print-ready / large-scale assets. - New model
recraft-v4-vector— $0.108 — native SVG output with editable paths and layers. The only generation models on the market shipping true vector files. - New model
recraft-v4-vector-pro— $0.405 — V4 vector at 4MP for print-ready logos and large-scale signage. - New model
flux-2-pro— 0.101 — BFL’s 32B flagship (3× larger than Flux 1.1). Photoreal, ~60% accurate text-in-image, unified gen+edit. Cheaper thanflux-1.1-ultraat 1MP. - New API param
image_urls: string[]— multi-reference blending for FLUX.2 Pro, up to 10 source images merged into a single output. - New per-megapixel pricing mode — FLUX.2 Pro bills a flat base for the first megapixel plus a per-extra-MP rate, rounded to the nearest whole MP. Hold uses the requested size; finalize uses the actual output dimensions.
recraft-v3andflux-1.1-ultraare now marked legacy. Existing API contracts continue to work; new projects should userecraft-v4andflux-2-pro.
April 29, 2026
Audio - 2 new endpoints + 2 SKUs
Kyma now hears. Two synchronous audio endpoints behind the same single-key gate as text, image, and video.- New endpoint
POST /v1/audio/transcriptions- speech-to-text, multipart upload, OpenAI Whisper API compatible - New endpoint
POST /v1/audio/understand- audio scene Q&A (tone, music, SFX, language, emotion), custom Kyma endpoint - New models:
whisper-v3-turboat 0.000648/min - Per-minute pricing - both endpoints bill in 1-minute increments, rounded up. 1-hour file: 0.039 understand = $0.093 total
- New aliases -
model: "transcribe"andmodel: "audio-understand"ride forward when underlying SKUs change - Audio rows now flow through the same V2 ledger as text - visible on
/logs,/rankings, and admin scorecards - Companion CLI
watch-cli- open-source orchestrator that gives any agent eyes and ears for any social video URL (~50x cheaper than full multimodal LLM analysis)
April 26, 2026
Video Generation - 5 new models
Five video models now live behind a single async endpoint.- New endpoint
POST /v1/videos/generations- async, returns 202 with ajob_id; pollGET /v1/jobs/{id}for the result - New models:
kling-2.5-pro,kling-3-pro,kling-3-pro-audio,seedance-2-pro,seedance-2-fast - Per-second pricing - 0.410 per second of video, billed against actual clamped duration
- Per-SKU duration caps - Kling 3 family and Seedance support up to 15s; Kling 2.5 stays at 10s
- Hold-and-finalize billing - failures refund in full; idempotency keys supported end-to-end
- T2V or I2V from a single endpoint - pass
image_urlto switch any video model into image-to-video mode
April 25, 2026
DeepSeek V4 — Pro and Flash
DeepSeek’s V4 lineup now live on Kyma. Both variants are MIT-licensed, MoE, with 1M context and native reasoning.deepseek-v4-pro— 1.6T (49B active) flagship for top reasoning and complex coding. 4.70 per 1M.deepseek-v4-flash— 284B (13B active) value tier. Same family behavior at the lowest V4 price. 0.38 per 1M.- 1M context window, 65K max output, tool calling and structured outputs supported on both.
deepseek-v3stays available as the previous-gen stable baseline; older workloads do not need to migrate.
Image Generation — Week 1
Four image-generation models now live behind a single async endpoint.- New endpoint
POST /v1/images/generations— async, returns 202 with ajob_id; pollGET /v1/jobs/{id}for the result - New models:
flux-1.1-ultra,flux-kontext-pro(image-edit),ideogram-v3,recraft-v3 - Pay-per-image billing with hold-and-finalize semantics — failures are refunded in full
- Idempotency key supported end-to-end; safe to retry POSTs without double-charging
April 23, 2026
API Reliability and Platform Changes
This release bundles billing and rate-limit changes, model/provider updates, agent/runtime improvements, dashboard and product polish, API behavior updates with a focus on user-visible improvements.- Per-model soft weight + latency tracking (#8)
- Ship Kimi K2.6 + fix kyma-agent auto-update EACCES (#9)
- /models nav link + response sanitization fix (#14)
- Routing — broadened fallback chain for deepseek-v3 (#7)
- Sync Kimi K2.6 across docs, landing pages, integrations (#11)
April 21, 2026
Product and Dashboard Updates
This release bundles billing and rate-limit changes, model/provider updates, agent/runtime improvements, dashboard and product polish, API behavior updates with a focus on user-visible improvements.- AI discoverability + models SEO + billing fixes (#3)
- Public model detail — tabbed code blocks like dashboard
- Tuned deepseek-v3 fallback chain for better latency
- AI agent discoverability + public /models SEO pages (#2)
- Admin_top_users RPC — power user analytics without row limit
April 19, 2026
Product and Dashboard Updates
This release bundles billing and rate-limit changes, model/provider updates, agent/runtime improvements, dashboard and product polish, API behavior updates with a focus on user-visible improvements.- Chat playground — searchable model picker, model ID copy, cost per message
- Billing page redesign — Anthropic-style single input, price breakdown, auto top-up inputs
- Billing page redesign — Anthropic-style layout
- Billing page — card editing, invoice links, tier display, unified layout
- Billing — inline card modal, invoice creation, estimated taxes, setup checkout
April 17, 2026
Agent, Install, and Runtime Improvements
This release bundles billing and rate-limit changes, model/provider updates, agent/runtime improvements, dashboard and product polish, API behavior updates with a focus on user-visible improvements.- Switch tier system from spend-based to deposit-based
- Added a new fallback layer for qwen-3.6-plus
- Correct pricing link to docs.kymaapi.com/pricing
- Bump kyma-agent to 0.1.6 and ter bootstrap to 0.1.1
- Qwen-3.6-plus routing tuned to serve the real closed-weight model
April 16, 2026
Kyma Agent v0.1.12 — KYMA.md context + MCP servers
- KYMA.md body injection — project rules now prepend to every turn’s system prompt, not just the initial one. Precedence:
~/.kyma/agent/KYMA.md→./KYMA.md→./KYMA.local.md, withCLAUDE.mdandAGENTS.mdas project fallbacks. Edits apply on the next turn — no restart needed. /init— scaffoldKYMA.mdin one step. Detects JS/TS, Python, Rust, Go, Ruby, PHP and their frameworks, then proposes a starter file with frontmatter (model, thinking) + Stack / Conventions / Key files / Agent behavior sections. Preview before write./mcp— Model Context Protocol servers — bring any MCP-compatible tool server into Kyma. Configure in~/.kyma/agent/mcp.json(user) or./.kyma/mcp.json(project). Subcommands:/mcpto list,/mcp enable <name>,/mcp disable <name>,/mcp test <name>. Tools register automatically at session start asmcp__<server>__<tool>./statusconsolidation — merged/doctor,/balance,/usageinto a single/status. It now shows account, credits, tier limits, lifetime spend, session totals, API latency, local diagnostics, and MCP health in one view.
GLM family from Z.AI
- 3 new models added, bringing the active catalog to 16 models:
glm-5.1— flagship long-running coding agent, repo-scale engineering, 203K contextglm-4.5-air— cheap agentic bulk tasks, 131K contextglm-4.7-flash— cheap long-context throughput, 203K context
- Implicit caching — 50% off on cache hits where the underlying infrastructure supports it.
- Kyma Agent v0.1.11 — the
/modelsslash command now lists GLM 5.1, GLM 4.5 Air, and GLM 4.7 Flash alongside the existing 9 curated models. - Auto-failover — every GLM model has multi-layer fallbacks so requests keep flowing even if a backend is unavailable.
April 15, 2026
Kyma Agent v0.1.8
- Fixed
/doctordiagnostics inkymaafter the0.1.7bridge release. kyma-terremains pinned to0.1.7; this was a package-only hotfix for the bundled ESM diagnostics path.
April 11, 2026
Kyma CLI v0.3
- Historical note: this entry describes the April 11 launch state. Kyma Agent and Kyma Ter have since been retired.
kymacommand — At that time, install was described asnpm install -g kyma-api, then just typekymato start an interactive chat session from your terminal. This CLI has since been retired.- Interactive model picker — Type
/modelin chat to browse active models with arrow keys. Or/model deepseek-r1to switch instantly. - Slash commands —
/model,/models,/balance,/clear,/help,/exit— manage your session without leaving the chat. - Pipe mode —
cat error.log | kyma "fix this"orgit diff | kyma "review this". Auto-detects non-TTY and outputs clean text. - Device code login —
kyma loginopens your browser, auto-fills the code, copies to clipboard. Supports Google OAuth. - JSON mode —
kyma models --jsonfor scripts and CI. Auto-quiet in non-TTY environments.
April 10, 2026
Higher Limits, Better Emails
- 2.5x higher token limits — Free tier now gets 200K tokens/minute (was 80K) and 30 RPM (was 20). Your coding agents can run longer sessions without hitting walls. Tier 1+ also increased proportionally.
- Transactional emails — You’ll now receive a welcome email on signup, a receipt after every purchase, and a heads-up when your balance is running low. Auto top-up failures also notify you immediately.
- Model expansion — Added MiniMax M2.7, Nemotron 3 Super, Step 3.5 Flash, GLM 4.5 Air, Gemma 4 26B MoE at that time. Model grid reorganized into Recommended / Coding & Agents / Fast & Long Context categories.
- Compare page — New
/comparepage with honest Kyma vs other LLM gateways and direct APIs comparison, including benchmark data and migration snippets. - Prompt caching as USP — 48% cache hit rate on heavy users, 23% average cost savings. Now highlighted on homepage.
April 8, 2026
Models & Pricing
- DeepSeek V3 + R1 — added as primary models with multi-layer fallbacks. DeepSeek V3 is GPT-5 class quality. DeepSeek R1 is a reasoning model 96% cheaper than o1.
- Expanded backend redundancy — additional infrastructure for DeepSeek, Llama, and Qwen models. Adds redundancy and lower latency.
- 4 models disabled — removed a low-yield Qwen 3 235B variant (high failure rate),
gpt-oss-20b,llama-3.1-8b,gemma-3-27b(superseded by better models). - Pricing audit — corrected 8 model prices so every listed price covers full multi-route redundancy. If a backup route is too expensive, we remove it rather than raise your price.
- 19 active models at that time with multi-layer infrastructure redundancy.
April 7, 2026
⚡ Reliability & Performance
- Auto-failover — if a model’s primary infrastructure is down, requests automatically retry on backup providers. Most failures are invisible to you.
- Faster responses — reduced internal overhead by ~250ms per request through smarter caching.
- Better rate limiting — rate limits are now shared across our infrastructure (no more inconsistent counts).
- Fallback headers — responses include
X-Kyma-Fallback: truewhen a backup was used, plusX-Kyma-Fallback-Layer: 1|2|3indicating how deep the fallback went. - More models available — expanded backup infrastructure means models stay available even during provider outages.
April 4, 2026
🚀 Launch
- Launch model set — Llama 3.3 70B, Llama 4 Scout, Qwen 3 32B/235B, Gemma 4, Kimi K2, GPT-OSS, Gemini
- Dashboard — API keys, usage stats, playground, model browser
- Google Sign-In — one-click login
- OpenAI compatible — drop-in replacement for any OpenAI SDK
- Tier-based rate limits: Tier 0 (free) = 30 RPM, up to Tier 4 = 300 RPM
- Streaming support
- Supabase PostgreSQL backend